1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>LLVM Assembly Language Reference Manual</title>
6 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
7 <meta name="author" content="Chris Lattner">
8 <meta name="description"
9 content="LLVM Assembly Language Reference Manual.">
10 <link rel="stylesheet" href="llvm.css" type="text/css">
15 <div class="doc_title"> LLVM Language Reference Manual </div>
17 <li><a href="#abstract">Abstract</a></li>
18 <li><a href="#introduction">Introduction</a></li>
19 <li><a href="#identifiers">Identifiers</a></li>
20 <li><a href="#highlevel">High Level Structure</a>
22 <li><a href="#modulestructure">Module Structure</a></li>
23 <li><a href="#linkage">Linkage Types</a></li>
24 <li><a href="#callingconv">Calling Conventions</a></li>
25 <li><a href="#globalvars">Global Variables</a></li>
26 <li><a href="#functionstructure">Functions</a></li>
27 <li><a href="#paramattrs">Parameter Attributes</a></li>
28 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
29 <li><a href="#datalayout">Data Layout</a></li>
32 <li><a href="#typesystem">Type System</a>
34 <li><a href="#t_primitive">Primitive Types</a>
36 <li><a href="#t_classifications">Type Classifications</a></li>
39 <li><a href="#t_derived">Derived Types</a>
41 <li><a href="#t_array">Array Type</a></li>
42 <li><a href="#t_function">Function Type</a></li>
43 <li><a href="#t_pointer">Pointer Type</a></li>
44 <li><a href="#t_struct">Structure Type</a></li>
45 <li><a href="#t_pstruct">Packed Structure Type</a></li>
46 <li><a href="#t_vector">Vector Type</a></li>
47 <li><a href="#t_opaque">Opaque Type</a></li>
52 <li><a href="#constants">Constants</a>
54 <li><a href="#simpleconstants">Simple Constants</a>
55 <li><a href="#aggregateconstants">Aggregate Constants</a>
56 <li><a href="#globalconstants">Global Variable and Function Addresses</a>
57 <li><a href="#undefvalues">Undefined Values</a>
58 <li><a href="#constantexprs">Constant Expressions</a>
61 <li><a href="#othervalues">Other Values</a>
63 <li><a href="#inlineasm">Inline Assembler Expressions</a>
66 <li><a href="#instref">Instruction Reference</a>
68 <li><a href="#terminators">Terminator Instructions</a>
70 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
71 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
72 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
73 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
74 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
75 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
78 <li><a href="#binaryops">Binary Operations</a>
80 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
81 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
82 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
83 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
84 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
85 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
86 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
87 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
88 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
91 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
93 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
94 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
95 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
96 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
97 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
98 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
101 <li><a href="#vectorops">Vector Operations</a>
103 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
104 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
105 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
108 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
110 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
111 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
112 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
113 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
114 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
115 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
118 <li><a href="#convertops">Conversion Operations</a>
120 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
121 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
122 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
123 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
124 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
125 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
126 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
127 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
128 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
129 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
130 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
131 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
133 <li><a href="#otherops">Other Operations</a>
135 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
136 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
137 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
138 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
139 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
140 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
145 <li><a href="#intrinsics">Intrinsic Functions</a>
147 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
149 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
150 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
151 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
154 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
156 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
157 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
158 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
161 <li><a href="#int_codegen">Code Generator Intrinsics</a>
163 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
164 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
165 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
166 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
167 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
168 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
169 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
172 <li><a href="#int_libc">Standard C Library Intrinsics</a>
174 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
175 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
176 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
177 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
178 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
181 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
183 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
184 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
185 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
186 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
187 <li><a href="#int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic </a></li>
188 <li><a href="#int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic </a></li>
191 <li><a href="#int_debugger">Debugger intrinsics</a></li>
192 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
197 <div class="doc_author">
198 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
199 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
202 <!-- *********************************************************************** -->
203 <div class="doc_section"> <a name="abstract">Abstract </a></div>
204 <!-- *********************************************************************** -->
206 <div class="doc_text">
207 <p>This document is a reference manual for the LLVM assembly language.
208 LLVM is an SSA based representation that provides type safety,
209 low-level operations, flexibility, and the capability of representing
210 'all' high-level languages cleanly. It is the common code
211 representation used throughout all phases of the LLVM compilation
215 <!-- *********************************************************************** -->
216 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
217 <!-- *********************************************************************** -->
219 <div class="doc_text">
221 <p>The LLVM code representation is designed to be used in three
222 different forms: as an in-memory compiler IR, as an on-disk bytecode
223 representation (suitable for fast loading by a Just-In-Time compiler),
224 and as a human readable assembly language representation. This allows
225 LLVM to provide a powerful intermediate representation for efficient
226 compiler transformations and analysis, while providing a natural means
227 to debug and visualize the transformations. The three different forms
228 of LLVM are all equivalent. This document describes the human readable
229 representation and notation.</p>
231 <p>The LLVM representation aims to be light-weight and low-level
232 while being expressive, typed, and extensible at the same time. It
233 aims to be a "universal IR" of sorts, by being at a low enough level
234 that high-level ideas may be cleanly mapped to it (similar to how
235 microprocessors are "universal IR's", allowing many source languages to
236 be mapped to them). By providing type information, LLVM can be used as
237 the target of optimizations: for example, through pointer analysis, it
238 can be proven that a C automatic variable is never accessed outside of
239 the current function... allowing it to be promoted to a simple SSA
240 value instead of a memory location.</p>
244 <!-- _______________________________________________________________________ -->
245 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
247 <div class="doc_text">
249 <p>It is important to note that this document describes 'well formed'
250 LLVM assembly language. There is a difference between what the parser
251 accepts and what is considered 'well formed'. For example, the
252 following instruction is syntactically okay, but not well formed:</p>
255 %x = <a href="#i_add">add</a> i32 1, %x
258 <p>...because the definition of <tt>%x</tt> does not dominate all of
259 its uses. The LLVM infrastructure provides a verification pass that may
260 be used to verify that an LLVM module is well formed. This pass is
261 automatically run by the parser after parsing input assembly and by
262 the optimizer before it outputs bytecode. The violations pointed out
263 by the verifier pass indicate bugs in transformation passes or input to
266 <!-- Describe the typesetting conventions here. --> </div>
268 <!-- *********************************************************************** -->
269 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
270 <!-- *********************************************************************** -->
272 <div class="doc_text">
274 <p>LLVM uses three different forms of identifiers, for different
278 <li>Named values are represented as a string of characters with a '%' prefix.
279 For example, %foo, %DivisionByZero, %a.really.long.identifier. The actual
280 regular expression used is '<tt>%[a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
281 Identifiers which require other characters in their names can be surrounded
282 with quotes. In this way, anything except a <tt>"</tt> character can be used
285 <li>Unnamed values are represented as an unsigned numeric value with a '%'
286 prefix. For example, %12, %2, %44.</li>
288 <li>Constants, which are described in a <a href="#constants">section about
289 constants</a>, below.</li>
292 <p>LLVM requires that values start with a '%' sign for two reasons: Compilers
293 don't need to worry about name clashes with reserved words, and the set of
294 reserved words may be expanded in the future without penalty. Additionally,
295 unnamed identifiers allow a compiler to quickly come up with a temporary
296 variable without having to avoid symbol table conflicts.</p>
298 <p>Reserved words in LLVM are very similar to reserved words in other
299 languages. There are keywords for different opcodes
300 ('<tt><a href="#i_add">add</a></tt>',
301 '<tt><a href="#i_bitcast">bitcast</a></tt>',
302 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
303 href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
304 and others. These reserved words cannot conflict with variable names, because
305 none of them start with a '%' character.</p>
307 <p>Here is an example of LLVM code to multiply the integer variable
308 '<tt>%X</tt>' by 8:</p>
313 %result = <a href="#i_mul">mul</a> i32 %X, 8
316 <p>After strength reduction:</p>
319 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
322 <p>And the hard way:</p>
325 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
326 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
327 %result = <a href="#i_add">add</a> i32 %1, %1
330 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
331 important lexical features of LLVM:</p>
335 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
338 <li>Unnamed temporaries are created when the result of a computation is not
339 assigned to a named value.</li>
341 <li>Unnamed temporaries are numbered sequentially</li>
345 <p>...and it also shows a convention that we follow in this document. When
346 demonstrating instructions, we will follow an instruction with a comment that
347 defines the type and name of value produced. Comments are shown in italic
352 <!-- *********************************************************************** -->
353 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
354 <!-- *********************************************************************** -->
356 <!-- ======================================================================= -->
357 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
360 <div class="doc_text">
362 <p>LLVM programs are composed of "Module"s, each of which is a
363 translation unit of the input programs. Each module consists of
364 functions, global variables, and symbol table entries. Modules may be
365 combined together with the LLVM linker, which merges function (and
366 global variable) definitions, resolves forward declarations, and merges
367 symbol table entries. Here is an example of the "hello world" module:</p>
369 <pre><i>; Declare the string constant as a global constant...</i>
370 <a href="#identifiers">%.LC0</a> = <a href="#linkage_internal">internal</a> <a
371 href="#globalvars">constant</a> <a href="#t_array">[13 x i8 ]</a> c"hello world\0A\00" <i>; [13 x i8 ]*</i>
373 <i>; External declaration of the puts function</i>
374 <a href="#functionstructure">declare</a> i32 %puts(i8 *) <i>; i32(i8 *)* </i>
376 <i>; Definition of main function</i>
377 define i32 %main() { <i>; i32()* </i>
378 <i>; Convert [13x i8 ]* to i8 *...</i>
380 href="#i_getelementptr">getelementptr</a> [13 x i8 ]* %.LC0, i64 0, i64 0 <i>; i8 *</i>
382 <i>; Call puts function to write out the string to stdout...</i>
384 href="#i_call">call</a> i32 %puts(i8 * %cast210) <i>; i32</i>
386 href="#i_ret">ret</a> i32 0<br>}<br></pre>
388 <p>This example is made up of a <a href="#globalvars">global variable</a>
389 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
390 function, and a <a href="#functionstructure">function definition</a>
391 for "<tt>main</tt>".</p>
393 <p>In general, a module is made up of a list of global values,
394 where both functions and global variables are global values. Global values are
395 represented by a pointer to a memory location (in this case, a pointer to an
396 array of char, and a pointer to a function), and have one of the following <a
397 href="#linkage">linkage types</a>.</p>
401 <!-- ======================================================================= -->
402 <div class="doc_subsection">
403 <a name="linkage">Linkage Types</a>
406 <div class="doc_text">
409 All Global Variables and Functions have one of the following types of linkage:
414 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
416 <dd>Global values with internal linkage are only directly accessible by
417 objects in the current module. In particular, linking code into a module with
418 an internal global value may cause the internal to be renamed as necessary to
419 avoid collisions. Because the symbol is internal to the module, all
420 references can be updated. This corresponds to the notion of the
421 '<tt>static</tt>' keyword in C.
424 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
426 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
427 the same name when linkage occurs. This is typically used to implement
428 inline functions, templates, or other code which must be generated in each
429 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
430 allowed to be discarded.
433 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
435 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
436 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
437 used for globals that may be emitted in multiple translation units, but that
438 are not guaranteed to be emitted into every translation unit that uses them.
439 One example of this are common globals in C, such as "<tt>int X;</tt>" at
443 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
445 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
446 pointer to array type. When two global variables with appending linkage are
447 linked together, the two global arrays are appended together. This is the
448 LLVM, typesafe, equivalent of having the system linker append together
449 "sections" with identical names when .o files are linked.
452 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
453 <dd>The semantics of this linkage follow the ELF model: the symbol is weak
454 until linked, if not linked, the symbol becomes null instead of being an
458 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
460 <dd>If none of the above identifiers are used, the global is externally
461 visible, meaning that it participates in linkage and can be used to resolve
462 external symbol references.
467 The next two types of linkage are targeted for Microsoft Windows platform
468 only. They are designed to support importing (exporting) symbols from (to)
473 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
475 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
476 or variable via a global pointer to a pointer that is set up by the DLL
477 exporting the symbol. On Microsoft Windows targets, the pointer name is
478 formed by combining <code>_imp__</code> and the function or variable name.
481 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
483 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
484 pointer to a pointer in a DLL, so that it can be referenced with the
485 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
486 name is formed by combining <code>_imp__</code> and the function or variable
492 <p><a name="linkage_external"></a>For example, since the "<tt>.LC0</tt>"
493 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
494 variable and was linked with this one, one of the two would be renamed,
495 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
496 external (i.e., lacking any linkage declarations), they are accessible
497 outside of the current module.</p>
498 <p>It is illegal for a function <i>declaration</i>
499 to have any linkage type other than "externally visible", <tt>dllimport</tt>,
500 or <tt>extern_weak</tt>.</p>
504 <!-- ======================================================================= -->
505 <div class="doc_subsection">
506 <a name="callingconv">Calling Conventions</a>
509 <div class="doc_text">
511 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
512 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
513 specified for the call. The calling convention of any pair of dynamic
514 caller/callee must match, or the behavior of the program is undefined. The
515 following calling conventions are supported by LLVM, and more may be added in
519 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
521 <dd>This calling convention (the default if no other calling convention is
522 specified) matches the target C calling conventions. This calling convention
523 supports varargs function calls and tolerates some mismatch in the declared
524 prototype and implemented declaration of the function (as does normal C).
527 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
529 <dd>This calling convention attempts to make calls as fast as possible
530 (e.g. by passing things in registers). This calling convention allows the
531 target to use whatever tricks it wants to produce fast code for the target,
532 without having to conform to an externally specified ABI. Implementations of
533 this convention should allow arbitrary tail call optimization to be supported.
534 This calling convention does not support varargs and requires the prototype of
535 all callees to exactly match the prototype of the function definition.
538 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
540 <dd>This calling convention attempts to make code in the caller as efficient
541 as possible under the assumption that the call is not commonly executed. As
542 such, these calls often preserve all registers so that the call does not break
543 any live ranges in the caller side. This calling convention does not support
544 varargs and requires the prototype of all callees to exactly match the
545 prototype of the function definition.
548 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
550 <dd>Any calling convention may be specified by number, allowing
551 target-specific calling conventions to be used. Target specific calling
552 conventions start at 64.
556 <p>More calling conventions can be added/defined on an as-needed basis, to
557 support pascal conventions or any other well-known target-independent
562 <!-- ======================================================================= -->
563 <div class="doc_subsection">
564 <a name="visibility">Visibility Styles</a>
567 <div class="doc_text">
570 All Global Variables and Functions have one of the following visibility styles:
574 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
576 <dd>On ELF, default visibility means that the declaration is visible to other
577 modules and, in shared libraries, means that the declared entity may be
578 overridden. On Darwin, default visibility means that the declaration is
579 visible to other modules. Default visibility corresponds to "external
580 linkage" in the language.
583 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
585 <dd>Two declarations of an object with hidden visibility refer to the same
586 object if they are in the same shared object. Usually, hidden visibility
587 indicates that the symbol will not be placed into the dynamic symbol table,
588 so no other module (executable or shared library) can reference it
596 <!-- ======================================================================= -->
597 <div class="doc_subsection">
598 <a name="globalvars">Global Variables</a>
601 <div class="doc_text">
603 <p>Global variables define regions of memory allocated at compilation time
604 instead of run-time. Global variables may optionally be initialized, may have
605 an explicit section to be placed in, and may
606 have an optional explicit alignment specified. A variable may be defined as
607 "thread_local", which means that it will not be shared by threads (each thread
608 will have a separated copy of the variable).
609 A variable may be defined as a global "constant," which indicates that the
610 contents of the variable will <b>never</b> be modified (enabling better
611 optimization, allowing the global data to be placed in the read-only section of
612 an executable, etc). Note that variables that need runtime initialization
613 cannot be marked "constant" as there is a store to the variable.</p>
616 LLVM explicitly allows <em>declarations</em> of global variables to be marked
617 constant, even if the final definition of the global is not. This capability
618 can be used to enable slightly better optimization of the program, but requires
619 the language definition to guarantee that optimizations based on the
620 'constantness' are valid for the translation units that do not include the
624 <p>As SSA values, global variables define pointer values that are in
625 scope (i.e. they dominate) all basic blocks in the program. Global
626 variables always define a pointer to their "content" type because they
627 describe a region of memory, and all memory objects in LLVM are
628 accessed through pointers.</p>
630 <p>LLVM allows an explicit section to be specified for globals. If the target
631 supports it, it will emit globals to the section specified.</p>
633 <p>An explicit alignment may be specified for a global. If not present, or if
634 the alignment is set to zero, the alignment of the global is set by the target
635 to whatever it feels convenient. If an explicit alignment is specified, the
636 global is forced to have at least that much alignment. All alignments must be
639 <p>For example, the following defines a global with an initializer, section,
643 %G = constant float 1.0, section "foo", align 4
649 <!-- ======================================================================= -->
650 <div class="doc_subsection">
651 <a name="functionstructure">Functions</a>
654 <div class="doc_text">
656 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
657 an optional <a href="#linkage">linkage type</a>, an optional
658 <a href="#visibility">visibility style</a>, an optional
659 <a href="#callingconv">calling convention</a>, a return type, an optional
660 <a href="#paramattrs">parameter attribute</a> for the return type, a function
661 name, a (possibly empty) argument list (each with optional
662 <a href="#paramattrs">parameter attributes</a>), an optional section, an
663 optional alignment, an opening curly brace, a list of basic blocks, and a
666 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
667 optional <a href="#linkage">linkage type</a>, an optional
668 <a href="#visibility">visibility style</a>, an optional
669 <a href="#callingconv">calling convention</a>, a return type, an optional
670 <a href="#paramattrs">parameter attribute</a> for the return type, a function
671 name, a possibly empty list of arguments, and an optional alignment.</p>
673 <p>A function definition contains a list of basic blocks, forming the CFG for
674 the function. Each basic block may optionally start with a label (giving the
675 basic block a symbol table entry), contains a list of instructions, and ends
676 with a <a href="#terminators">terminator</a> instruction (such as a branch or
677 function return).</p>
679 <p>The first basic block in a program is special in two ways: it is immediately
680 executed on entrance to the function, and it is not allowed to have predecessor
681 basic blocks (i.e. there can not be any branches to the entry block of a
682 function). Because the block can have no predecessors, it also cannot have any
683 <a href="#i_phi">PHI nodes</a>.</p>
685 <p>LLVM functions are identified by their name and type signature. Hence, two
686 functions with the same name but different parameter lists or return values are
687 considered different functions, and LLVM will resolve references to each
690 <p>LLVM allows an explicit section to be specified for functions. If the target
691 supports it, it will emit functions to the section specified.</p>
693 <p>An explicit alignment may be specified for a function. If not present, or if
694 the alignment is set to zero, the alignment of the function is set by the target
695 to whatever it feels convenient. If an explicit alignment is specified, the
696 function is forced to have at least that much alignment. All alignments must be
701 <!-- ======================================================================= -->
702 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
703 <div class="doc_text">
704 <p>The return type and each parameter of a function type may have a set of
705 <i>parameter attributes</i> associated with them. Parameter attributes are
706 used to communicate additional information about the result or parameters of
707 a function. Parameter attributes are considered to be part of the function
708 type so two functions types that differ only by the parameter attributes
709 are different function types.</p>
711 <p>Parameter attributes are simple keywords that follow the type specified. If
712 multiple parameter attributes are needed, they are space separated. For
714 %someFunc = i16 (i8 sext %someParam) zext
715 %someFunc = i16 (i8 zext %someParam) zext</pre>
716 <p>Note that the two function types above are unique because the parameter has
717 a different attribute (sext in the first one, zext in the second). Also note
718 that the attribute for the function result (zext) comes immediately after the
721 <p>Currently, only the following parameter attributes are defined:</p>
723 <dt><tt>zext</tt></dt>
724 <dd>This indicates that the parameter should be zero extended just before
725 a call to this function.</dd>
726 <dt><tt>sext</tt></dt>
727 <dd>This indicates that the parameter should be sign extended just before
728 a call to this function.</dd>
729 <dt><tt>inreg</tt></dt>
730 <dd>This indicates that the parameter should be placed in register (if
731 possible) during assembling function call. Support for this attribute is
733 <dt><tt>sret</tt></dt>
734 <dd>This indicates that the parameter specifies the address of a structure
735 that is the return value of the function in the source program.</dd>
736 <dt><tt>noreturn</tt></dt>
737 <dd>This function attribute indicates that the function never returns. This
738 indicates to LLVM that every call to this function should be treated as if
739 an <tt>unreachable</tt> instruction immediately followed the call.</dd>
740 <dt><tt>nounwind</tt></dt>
741 <dd>This function attribute indicates that the function type does not use
742 the unwind instruction and does not allow stack unwinding to propagate
748 <!-- ======================================================================= -->
749 <div class="doc_subsection">
750 <a name="moduleasm">Module-Level Inline Assembly</a>
753 <div class="doc_text">
755 Modules may contain "module-level inline asm" blocks, which corresponds to the
756 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
757 LLVM and treated as a single unit, but may be separated in the .ll file if
758 desired. The syntax is very simple:
761 <div class="doc_code"><pre>
762 module asm "inline asm code goes here"
763 module asm "more can go here"
766 <p>The strings can contain any character by escaping non-printable characters.
767 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
772 The inline asm code is simply printed to the machine code .s file when
773 assembly code is generated.
777 <!-- ======================================================================= -->
778 <div class="doc_subsection">
779 <a name="datalayout">Data Layout</a>
782 <div class="doc_text">
783 <p>A module may specify a target specific data layout string that specifies how
784 data is to be laid out in memory. The syntax for the data layout is simply:</p>
785 <pre> target datalayout = "<i>layout specification</i>"</pre>
786 <p>The <i>layout specification</i> consists of a list of specifications
787 separated by the minus sign character ('-'). Each specification starts with a
788 letter and may include other information after the letter to define some
789 aspect of the data layout. The specifications accepted are as follows: </p>
792 <dd>Specifies that the target lays out data in big-endian form. That is, the
793 bits with the most significance have the lowest address location.</dd>
795 <dd>Specifies that hte target lays out data in little-endian form. That is,
796 the bits with the least significance have the lowest address location.</dd>
797 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
798 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
799 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
800 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
802 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
803 <dd>This specifies the alignment for an integer type of a given bit
804 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
805 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
806 <dd>This specifies the alignment for a vector type of a given bit
808 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
809 <dd>This specifies the alignment for a floating point type of a given bit
810 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
812 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
813 <dd>This specifies the alignment for an aggregate type of a given bit
816 <p>When constructing the data layout for a given target, LLVM starts with a
817 default set of specifications which are then (possibly) overriden by the
818 specifications in the <tt>datalayout</tt> keyword. The default specifications
819 are given in this list:</p>
821 <li><tt>E</tt> - big endian</li>
822 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
823 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
824 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
825 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
826 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
827 <li><tt>i64:32:64</tt> - i64 has abi alignment of 32-bits but preferred
828 alignment of 64-bits</li>
829 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
830 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
831 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
832 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
833 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
835 <p>When llvm is determining the alignment for a given type, it uses the
838 <li>If the type sought is an exact match for one of the specifications, that
839 specification is used.</li>
840 <li>If no match is found, and the type sought is an integer type, then the
841 smallest integer type that is larger than the bitwidth of the sought type is
842 used. If none of the specifications are larger than the bitwidth then the the
843 largest integer type is used. For example, given the default specifications
844 above, the i7 type will use the alignment of i8 (next largest) while both
845 i65 and i256 will use the alignment of i64 (largest specified).</li>
846 <li>If no match is found, and the type sought is a vector type, then the
847 largest vector type that is smaller than the sought vector type will be used
848 as a fall back. This happens because <128 x double> can be implemented in
849 terms of 64 <2 x double>, for example.</li>
853 <!-- *********************************************************************** -->
854 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
855 <!-- *********************************************************************** -->
857 <div class="doc_text">
859 <p>The LLVM type system is one of the most important features of the
860 intermediate representation. Being typed enables a number of
861 optimizations to be performed on the IR directly, without having to do
862 extra analyses on the side before the transformation. A strong type
863 system makes it easier to read the generated code and enables novel
864 analyses and transformations that are not feasible to perform on normal
865 three address code representations.</p>
869 <!-- ======================================================================= -->
870 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
871 <div class="doc_text">
872 <p>The primitive types are the fundamental building blocks of the LLVM
873 system. The current set of primitive types is as follows:</p>
875 <table class="layout">
880 <tr><th>Type</th><th>Description</th></tr>
881 <tr><td><tt><a name="t_void">void</a></tt></td><td>No value</td></tr>
882 <tr><td><tt>i8</tt></td><td>8-bit value</td></tr>
883 <tr><td><tt>i32</tt></td><td>32-bit value</td></tr>
884 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
885 <tr><td><tt>label</tt></td><td>Branch destination</td></tr>
892 <tr><th>Type</th><th>Description</th></tr>
893 <tr><td><tt>i1</tt></td><td>True or False value</td></tr>
894 <tr><td><tt>i16</tt></td><td>16-bit value</td></tr>
895 <tr><td><tt>i64</tt></td><td>64-bit value</td></tr>
896 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
904 <!-- _______________________________________________________________________ -->
905 <div class="doc_subsubsection"> <a name="t_classifications">Type
906 Classifications</a> </div>
907 <div class="doc_text">
908 <p>These different primitive types fall into a few useful
911 <table border="1" cellspacing="0" cellpadding="4">
913 <tr><th>Classification</th><th>Types</th></tr>
915 <td><a name="t_integer">integer</a></td>
916 <td><tt>i1, i8, i16, i32, i64</tt></td>
919 <td><a name="t_floating">floating point</a></td>
920 <td><tt>float, double</tt></td>
923 <td><a name="t_firstclass">first class</a></td>
924 <td><tt>i1, i8, i16, i32, i64, float, double, <br/>
925 <a href="#t_pointer">pointer</a>,<a href="#t_vector">vector</a></tt>
931 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
932 most important. Values of these types are the only ones which can be
933 produced by instructions, passed as arguments, or used as operands to
934 instructions. This means that all structures and arrays must be
935 manipulated either by pointer or by component.</p>
938 <!-- ======================================================================= -->
939 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
941 <div class="doc_text">
943 <p>The real power in LLVM comes from the derived types in the system.
944 This is what allows a programmer to represent arrays, functions,
945 pointers, and other useful types. Note that these derived types may be
946 recursive: For example, it is possible to have a two dimensional array.</p>
950 <!-- _______________________________________________________________________ -->
951 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
953 <div class="doc_text">
957 <p>The array type is a very simple derived type that arranges elements
958 sequentially in memory. The array type requires a size (number of
959 elements) and an underlying data type.</p>
964 [<# elements> x <elementtype>]
967 <p>The number of elements is a constant integer value; elementtype may
968 be any type with a size.</p>
971 <table class="layout">
974 <tt>[40 x i32 ]</tt><br/>
975 <tt>[41 x i32 ]</tt><br/>
976 <tt>[40 x i8]</tt><br/>
979 Array of 40 32-bit integer values.<br/>
980 Array of 41 32-bit integer values.<br/>
981 Array of 40 8-bit integer values.<br/>
985 <p>Here are some examples of multidimensional arrays:</p>
986 <table class="layout">
989 <tt>[3 x [4 x i32]]</tt><br/>
990 <tt>[12 x [10 x float]]</tt><br/>
991 <tt>[2 x [3 x [4 x i16]]]</tt><br/>
994 3x4 array of 32-bit integer values.<br/>
995 12x10 array of single precision floating point values.<br/>
996 2x3x4 array of 16-bit integer values.<br/>
1001 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1002 length array. Normally, accesses past the end of an array are undefined in
1003 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1004 As a special case, however, zero length arrays are recognized to be variable
1005 length. This allows implementation of 'pascal style arrays' with the LLVM
1006 type "{ i32, [0 x float]}", for example.</p>
1010 <!-- _______________________________________________________________________ -->
1011 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1012 <div class="doc_text">
1014 <p>The function type can be thought of as a function signature. It
1015 consists of a return type and a list of formal parameter types.
1016 Function types are usually used to build virtual function tables
1017 (which are structures of pointers to functions), for indirect function
1018 calls, and when defining a function.</p>
1020 The return type of a function type cannot be an aggregate type.
1023 <pre> <returntype> (<parameter list>)<br></pre>
1024 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1025 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1026 which indicates that the function takes a variable number of arguments.
1027 Variable argument functions can access their arguments with the <a
1028 href="#int_varargs">variable argument handling intrinsic</a> functions.</p>
1030 <table class="layout">
1032 <td class="left"><tt>i32 (i32)</tt></td>
1033 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1035 </tr><tr class="layout">
1036 <td class="left"><tt>float (i16 sext, i32 *) *
1038 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1039 an <tt>i16</tt> that should be sign extended and a
1040 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1043 </tr><tr class="layout">
1044 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1045 <td class="left">A vararg function that takes at least one
1046 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1047 which returns an integer. This is the signature for <tt>printf</tt> in
1054 <!-- _______________________________________________________________________ -->
1055 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1056 <div class="doc_text">
1058 <p>The structure type is used to represent a collection of data members
1059 together in memory. The packing of the field types is defined to match
1060 the ABI of the underlying processor. The elements of a structure may
1061 be any type that has a size.</p>
1062 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1063 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1064 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1067 <pre> { <type list> }<br></pre>
1069 <table class="layout">
1072 <tt>{ i32, i32, i32 }</tt><br/>
1073 <tt>{ float, i32 (i32) * }</tt><br/>
1076 a triple of three <tt>i32</tt> values<br/>
1077 A pair, where the first element is a <tt>float</tt> and the second element
1078 is a <a href="#t_pointer">pointer</a> to a <a href="#t_function">function</a>
1079 that takes an <tt>i32</tt>, returning an <tt>i32</tt>.<br/>
1085 <!-- _______________________________________________________________________ -->
1086 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1088 <div class="doc_text">
1090 <p>The packed structure type is used to represent a collection of data members
1091 together in memory. There is no padding between fields. Further, the alignment
1092 of a packed structure is 1 byte. The elements of a packed structure may
1093 be any type that has a size.</p>
1094 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1095 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1096 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1099 <pre> < { <type list> } > <br></pre>
1101 <table class="layout">
1104 <tt> < { i32, i32, i32 } > </tt><br/>
1105 <tt> < { float, i32 (i32) * } > </tt><br/>
1108 a triple of three <tt>i32</tt> values<br/>
1109 A pair, where the first element is a <tt>float</tt> and the second element
1110 is a <a href="#t_pointer">pointer</a> to a <a href="#t_function">function</a>
1111 that takes an <tt>i32</tt>, returning an <tt>i32</tt>.<br/>
1117 <!-- _______________________________________________________________________ -->
1118 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1119 <div class="doc_text">
1121 <p>As in many languages, the pointer type represents a pointer or
1122 reference to another object, which must live in memory.</p>
1124 <pre> <type> *<br></pre>
1126 <table class="layout">
1129 <tt>[4x i32]*</tt><br/>
1130 <tt>i32 (i32 *) *</tt><br/>
1133 A <a href="#t_pointer">pointer</a> to <a href="#t_array">array</a> of
1134 four <tt>i32</tt> values<br/>
1135 A <a href="#t_pointer">pointer</a> to a <a
1136 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1143 <!-- _______________________________________________________________________ -->
1144 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1145 <div class="doc_text">
1149 <p>A vector type is a simple derived type that represents a vector
1150 of elements. Vector types are used when multiple primitive data
1151 are operated in parallel using a single instruction (SIMD).
1152 A vector type requires a size (number of
1153 elements) and an underlying primitive data type. Vectors must have a power
1154 of two length (1, 2, 4, 8, 16 ...). Vector types are
1155 considered <a href="#t_firstclass">first class</a>.</p>
1160 < <# elements> x <elementtype> >
1163 <p>The number of elements is a constant integer value; elementtype may
1164 be any integer or floating point type.</p>
1168 <table class="layout">
1171 <tt><4 x i32></tt><br/>
1172 <tt><8 x float></tt><br/>
1173 <tt><2 x i64></tt><br/>
1176 Vector of 4 32-bit integer values.<br/>
1177 Vector of 8 floating-point values.<br/>
1178 Vector of 2 64-bit integer values.<br/>
1184 <!-- _______________________________________________________________________ -->
1185 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1186 <div class="doc_text">
1190 <p>Opaque types are used to represent unknown types in the system. This
1191 corresponds (for example) to the C notion of a foward declared structure type.
1192 In LLVM, opaque types can eventually be resolved to any type (not just a
1193 structure type).</p>
1203 <table class="layout">
1209 An opaque type.<br/>
1216 <!-- *********************************************************************** -->
1217 <div class="doc_section"> <a name="constants">Constants</a> </div>
1218 <!-- *********************************************************************** -->
1220 <div class="doc_text">
1222 <p>LLVM has several different basic types of constants. This section describes
1223 them all and their syntax.</p>
1227 <!-- ======================================================================= -->
1228 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1230 <div class="doc_text">
1233 <dt><b>Boolean constants</b></dt>
1235 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1236 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1239 <dt><b>Integer constants</b></dt>
1241 <dd>Standard integers (such as '4') are constants of the <a
1242 href="#t_integer">integer</a> type. Negative numbers may be used with
1246 <dt><b>Floating point constants</b></dt>
1248 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1249 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1250 notation (see below). Floating point constants must have a <a
1251 href="#t_floating">floating point</a> type. </dd>
1253 <dt><b>Null pointer constants</b></dt>
1255 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1256 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1260 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1261 of floating point constants. For example, the form '<tt>double
1262 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1263 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1264 (and the only time that they are generated by the disassembler) is when a
1265 floating point constant must be emitted but it cannot be represented as a
1266 decimal floating point number. For example, NaN's, infinities, and other
1267 special values are represented in their IEEE hexadecimal format so that
1268 assembly and disassembly do not cause any bits to change in the constants.</p>
1272 <!-- ======================================================================= -->
1273 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1276 <div class="doc_text">
1277 <p>Aggregate constants arise from aggregation of simple constants
1278 and smaller aggregate constants.</p>
1281 <dt><b>Structure constants</b></dt>
1283 <dd>Structure constants are represented with notation similar to structure
1284 type definitions (a comma separated list of elements, surrounded by braces
1285 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* %G }</tt>",
1286 where "<tt>%G</tt>" is declared as "<tt>%G = external global i32</tt>". Structure constants
1287 must have <a href="#t_struct">structure type</a>, and the number and
1288 types of elements must match those specified by the type.
1291 <dt><b>Array constants</b></dt>
1293 <dd>Array constants are represented with notation similar to array type
1294 definitions (a comma separated list of elements, surrounded by square brackets
1295 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1296 constants must have <a href="#t_array">array type</a>, and the number and
1297 types of elements must match those specified by the type.
1300 <dt><b>Vector constants</b></dt>
1302 <dd>Vector constants are represented with notation similar to vector type
1303 definitions (a comma separated list of elements, surrounded by
1304 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1305 i32 11, i32 74, i32 100 ></tt>". VEctor constants must have <a
1306 href="#t_vector">vector type</a>, and the number and types of elements must
1307 match those specified by the type.
1310 <dt><b>Zero initialization</b></dt>
1312 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1313 value to zero of <em>any</em> type, including scalar and aggregate types.
1314 This is often used to avoid having to print large zero initializers (e.g. for
1315 large arrays) and is always exactly equivalent to using explicit zero
1322 <!-- ======================================================================= -->
1323 <div class="doc_subsection">
1324 <a name="globalconstants">Global Variable and Function Addresses</a>
1327 <div class="doc_text">
1329 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1330 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1331 constants. These constants are explicitly referenced when the <a
1332 href="#identifiers">identifier for the global</a> is used and always have <a
1333 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1339 %Z = global [2 x i32*] [ i32* %X, i32* %Y ]
1344 <!-- ======================================================================= -->
1345 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1346 <div class="doc_text">
1347 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1348 no specific value. Undefined values may be of any type and be used anywhere
1349 a constant is permitted.</p>
1351 <p>Undefined values indicate to the compiler that the program is well defined
1352 no matter what value is used, giving the compiler more freedom to optimize.
1356 <!-- ======================================================================= -->
1357 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1360 <div class="doc_text">
1362 <p>Constant expressions are used to allow expressions involving other constants
1363 to be used as constants. Constant expressions may be of any <a
1364 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1365 that does not have side effects (e.g. load and call are not supported). The
1366 following is the syntax for constant expressions:</p>
1369 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1370 <dd>Truncate a constant to another type. The bit size of CST must be larger
1371 than the bit size of TYPE. Both types must be integers.</dd>
1373 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1374 <dd>Zero extend a constant to another type. The bit size of CST must be
1375 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1377 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1378 <dd>Sign extend a constant to another type. The bit size of CST must be
1379 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1381 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1382 <dd>Truncate a floating point constant to another floating point type. The
1383 size of CST must be larger than the size of TYPE. Both types must be
1384 floating point.</dd>
1386 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1387 <dd>Floating point extend a constant to another type. The size of CST must be
1388 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1390 <dt><b><tt>fp2uint ( CST to TYPE )</tt></b></dt>
1391 <dd>Convert a floating point constant to the corresponding unsigned integer
1392 constant. TYPE must be an integer type. CST must be floating point. If the
1393 value won't fit in the integer type, the results are undefined.</dd>
1395 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1396 <dd>Convert a floating point constant to the corresponding signed integer
1397 constant. TYPE must be an integer type. CST must be floating point. If the
1398 value won't fit in the integer type, the results are undefined.</dd>
1400 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1401 <dd>Convert an unsigned integer constant to the corresponding floating point
1402 constant. TYPE must be floating point. CST must be of integer type. If the
1403 value won't fit in the floating point type, the results are undefined.</dd>
1405 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1406 <dd>Convert a signed integer constant to the corresponding floating point
1407 constant. TYPE must be floating point. CST must be of integer type. If the
1408 value won't fit in the floating point type, the results are undefined.</dd>
1410 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1411 <dd>Convert a pointer typed constant to the corresponding integer constant
1412 TYPE must be an integer type. CST must be of pointer type. The CST value is
1413 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1415 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1416 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1417 pointer type. CST must be of integer type. The CST value is zero extended,
1418 truncated, or unchanged to make it fit in a pointer size. This one is
1419 <i>really</i> dangerous!</dd>
1421 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1422 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1423 identical (same number of bits). The conversion is done as if the CST value
1424 was stored to memory and read back as TYPE. In other words, no bits change
1425 with this operator, just the type. This can be used for conversion of
1426 vector types to any other type, as long as they have the same bit width. For
1427 pointers it is only valid to cast to another pointer type.
1430 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1432 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1433 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1434 instruction, the index list may have zero or more indexes, which are required
1435 to make sense for the type of "CSTPTR".</dd>
1437 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1439 <dd>Perform the <a href="#i_select">select operation</a> on
1442 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1443 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1445 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1446 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1448 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1450 <dd>Perform the <a href="#i_extractelement">extractelement
1451 operation</a> on constants.
1453 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1455 <dd>Perform the <a href="#i_insertelement">insertelement
1456 operation</a> on constants.</dd>
1459 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1461 <dd>Perform the <a href="#i_shufflevector">shufflevector
1462 operation</a> on constants.</dd>
1464 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1466 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1467 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1468 binary</a> operations. The constraints on operands are the same as those for
1469 the corresponding instruction (e.g. no bitwise operations on floating point
1470 values are allowed).</dd>
1474 <!-- *********************************************************************** -->
1475 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1476 <!-- *********************************************************************** -->
1478 <!-- ======================================================================= -->
1479 <div class="doc_subsection">
1480 <a name="inlineasm">Inline Assembler Expressions</a>
1483 <div class="doc_text">
1486 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1487 Module-Level Inline Assembly</a>) through the use of a special value. This
1488 value represents the inline assembler as a string (containing the instructions
1489 to emit), a list of operand constraints (stored as a string), and a flag that
1490 indicates whether or not the inline asm expression has side effects. An example
1491 inline assembler expression is:
1495 i32 (i32) asm "bswap $0", "=r,r"
1499 Inline assembler expressions may <b>only</b> be used as the callee operand of
1500 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1504 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1508 Inline asms with side effects not visible in the constraint list must be marked
1509 as having side effects. This is done through the use of the
1510 '<tt>sideeffect</tt>' keyword, like so:
1514 call void asm sideeffect "eieio", ""()
1517 <p>TODO: The format of the asm and constraints string still need to be
1518 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1519 need to be documented).
1524 <!-- *********************************************************************** -->
1525 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1526 <!-- *********************************************************************** -->
1528 <div class="doc_text">
1530 <p>The LLVM instruction set consists of several different
1531 classifications of instructions: <a href="#terminators">terminator
1532 instructions</a>, <a href="#binaryops">binary instructions</a>,
1533 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1534 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1535 instructions</a>.</p>
1539 <!-- ======================================================================= -->
1540 <div class="doc_subsection"> <a name="terminators">Terminator
1541 Instructions</a> </div>
1543 <div class="doc_text">
1545 <p>As mentioned <a href="#functionstructure">previously</a>, every
1546 basic block in a program ends with a "Terminator" instruction, which
1547 indicates which block should be executed after the current block is
1548 finished. These terminator instructions typically yield a '<tt>void</tt>'
1549 value: they produce control flow, not values (the one exception being
1550 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1551 <p>There are six different terminator instructions: the '<a
1552 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1553 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1554 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1555 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1556 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1560 <!-- _______________________________________________________________________ -->
1561 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1562 Instruction</a> </div>
1563 <div class="doc_text">
1565 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1566 ret void <i>; Return from void function</i>
1569 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1570 value) from a function back to the caller.</p>
1571 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1572 returns a value and then causes control flow, and one that just causes
1573 control flow to occur.</p>
1575 <p>The '<tt>ret</tt>' instruction may return any '<a
1576 href="#t_firstclass">first class</a>' type. Notice that a function is
1577 not <a href="#wellformed">well formed</a> if there exists a '<tt>ret</tt>'
1578 instruction inside of the function that returns a value that does not
1579 match the return type of the function.</p>
1581 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1582 returns back to the calling function's context. If the caller is a "<a
1583 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1584 the instruction after the call. If the caller was an "<a
1585 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1586 at the beginning of the "normal" destination block. If the instruction
1587 returns a value, that value shall set the call or invoke instruction's
1590 <pre> ret i32 5 <i>; Return an integer value of 5</i>
1591 ret void <i>; Return from a void function</i>
1594 <!-- _______________________________________________________________________ -->
1595 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1596 <div class="doc_text">
1598 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1601 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1602 transfer to a different basic block in the current function. There are
1603 two forms of this instruction, corresponding to a conditional branch
1604 and an unconditional branch.</p>
1606 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1607 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
1608 unconditional form of the '<tt>br</tt>' instruction takes a single
1609 '<tt>label</tt>' value as a target.</p>
1611 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
1612 argument is evaluated. If the value is <tt>true</tt>, control flows
1613 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1614 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1616 <pre>Test:<br> %cond = <a href="#i_icmp">icmp</a> eq, i32 %a, %b<br> br i1 %cond, label %IfEqual, label %IfUnequal<br>IfEqual:<br> <a
1617 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1619 <!-- _______________________________________________________________________ -->
1620 <div class="doc_subsubsection">
1621 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1624 <div class="doc_text">
1628 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1633 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1634 several different places. It is a generalization of the '<tt>br</tt>'
1635 instruction, allowing a branch to occur to one of many possible
1641 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1642 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1643 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1644 table is not allowed to contain duplicate constant entries.</p>
1648 <p>The <tt>switch</tt> instruction specifies a table of values and
1649 destinations. When the '<tt>switch</tt>' instruction is executed, this
1650 table is searched for the given value. If the value is found, control flow is
1651 transfered to the corresponding destination; otherwise, control flow is
1652 transfered to the default destination.</p>
1654 <h5>Implementation:</h5>
1656 <p>Depending on properties of the target machine and the particular
1657 <tt>switch</tt> instruction, this instruction may be code generated in different
1658 ways. For example, it could be generated as a series of chained conditional
1659 branches or with a lookup table.</p>
1664 <i>; Emulate a conditional br instruction</i>
1665 %Val = <a href="#i_zext">zext</a> i1 %value to i32
1666 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
1668 <i>; Emulate an unconditional br instruction</i>
1669 switch i32 0, label %dest [ ]
1671 <i>; Implement a jump table:</i>
1672 switch i32 %val, label %otherwise [ i32 0, label %onzero
1674 i32 2, label %ontwo ]
1678 <!-- _______________________________________________________________________ -->
1679 <div class="doc_subsubsection">
1680 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1683 <div class="doc_text">
1688 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1689 to label <normal label> unwind label <exception label>
1694 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1695 function, with the possibility of control flow transfer to either the
1696 '<tt>normal</tt>' label or the
1697 '<tt>exception</tt>' label. If the callee function returns with the
1698 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1699 "normal" label. If the callee (or any indirect callees) returns with the "<a
1700 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1701 continued at the dynamically nearest "exception" label.</p>
1705 <p>This instruction requires several arguments:</p>
1709 The optional "cconv" marker indicates which <a href="#callingconv">calling
1710 convention</a> the call should use. If none is specified, the call defaults
1711 to using C calling conventions.
1713 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1714 function value being invoked. In most cases, this is a direct function
1715 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1716 an arbitrary pointer to function value.
1719 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1720 function to be invoked. </li>
1722 <li>'<tt>function args</tt>': argument list whose types match the function
1723 signature argument types. If the function signature indicates the function
1724 accepts a variable number of arguments, the extra arguments can be
1727 <li>'<tt>normal label</tt>': the label reached when the called function
1728 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
1730 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1731 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
1737 <p>This instruction is designed to operate as a standard '<tt><a
1738 href="#i_call">call</a></tt>' instruction in most regards. The primary
1739 difference is that it establishes an association with a label, which is used by
1740 the runtime library to unwind the stack.</p>
1742 <p>This instruction is used in languages with destructors to ensure that proper
1743 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
1744 exception. Additionally, this is important for implementation of
1745 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
1749 %retval = invoke i32 %Test(i32 15) to label %Continue
1750 unwind label %TestCleanup <i>; {i32}:retval set</i>
1751 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Test(i32 15) to label %Continue
1752 unwind label %TestCleanup <i>; {i32}:retval set</i>
1757 <!-- _______________________________________________________________________ -->
1759 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
1760 Instruction</a> </div>
1762 <div class="doc_text">
1771 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
1772 at the first callee in the dynamic call stack which used an <a
1773 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
1774 primarily used to implement exception handling.</p>
1778 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
1779 immediately halt. The dynamic call stack is then searched for the first <a
1780 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
1781 execution continues at the "exceptional" destination block specified by the
1782 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
1783 dynamic call chain, undefined behavior results.</p>
1786 <!-- _______________________________________________________________________ -->
1788 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
1789 Instruction</a> </div>
1791 <div class="doc_text">
1800 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
1801 instruction is used to inform the optimizer that a particular portion of the
1802 code is not reachable. This can be used to indicate that the code after a
1803 no-return function cannot be reached, and other facts.</p>
1807 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
1812 <!-- ======================================================================= -->
1813 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
1814 <div class="doc_text">
1815 <p>Binary operators are used to do most of the computation in a
1816 program. They require two operands, execute an operation on them, and
1817 produce a single value. The operands might represent
1818 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
1819 The result value of a binary operator is not
1820 necessarily the same type as its operands.</p>
1821 <p>There are several different binary operators:</p>
1823 <!-- _______________________________________________________________________ -->
1824 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
1825 Instruction</a> </div>
1826 <div class="doc_text">
1828 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1831 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
1833 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
1834 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
1835 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
1836 Both arguments must have identical types.</p>
1838 <p>The value produced is the integer or floating point sum of the two
1841 <pre> <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
1844 <!-- _______________________________________________________________________ -->
1845 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
1846 Instruction</a> </div>
1847 <div class="doc_text">
1849 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1852 <p>The '<tt>sub</tt>' instruction returns the difference of its two
1854 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
1855 instruction present in most other intermediate representations.</p>
1857 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
1858 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1860 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
1861 Both arguments must have identical types.</p>
1863 <p>The value produced is the integer or floating point difference of
1864 the two operands.</p>
1866 <pre> <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
1867 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
1870 <!-- _______________________________________________________________________ -->
1871 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
1872 Instruction</a> </div>
1873 <div class="doc_text">
1875 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1878 <p>The '<tt>mul</tt>' instruction returns the product of its two
1881 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
1882 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1884 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
1885 Both arguments must have identical types.</p>
1887 <p>The value produced is the integer or floating point product of the
1889 <p>Because the operands are the same width, the result of an integer
1890 multiplication is the same whether the operands should be deemed unsigned or
1893 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
1896 <!-- _______________________________________________________________________ -->
1897 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
1899 <div class="doc_text">
1901 <pre> <result> = udiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1904 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
1907 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
1908 <a href="#t_integer">integer</a> values. Both arguments must have identical
1909 types. This instruction can also take <a href="#t_vector">vector</a> versions
1910 of the values in which case the elements must be integers.</p>
1912 <p>The value produced is the unsigned integer quotient of the two operands. This
1913 instruction always performs an unsigned division operation, regardless of
1914 whether the arguments are unsigned or not.</p>
1916 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
1919 <!-- _______________________________________________________________________ -->
1920 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
1922 <div class="doc_text">
1924 <pre> <result> = sdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1927 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
1930 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
1931 <a href="#t_integer">integer</a> values. Both arguments must have identical
1932 types. This instruction can also take <a href="#t_vector">vector</a> versions
1933 of the values in which case the elements must be integers.</p>
1935 <p>The value produced is the signed integer quotient of the two operands. This
1936 instruction always performs a signed division operation, regardless of whether
1937 the arguments are signed or not.</p>
1939 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
1942 <!-- _______________________________________________________________________ -->
1943 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
1944 Instruction</a> </div>
1945 <div class="doc_text">
1947 <pre> <result> = fdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1950 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
1953 <p>The two arguments to the '<tt>div</tt>' instruction must be
1954 <a href="#t_floating">floating point</a> values. Both arguments must have
1955 identical types. This instruction can also take <a href="#t_vector">vector</a>
1956 versions of the values in which case the elements must be floating point.</p>
1958 <p>The value produced is the floating point quotient of the two operands.</p>
1960 <pre> <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
1963 <!-- _______________________________________________________________________ -->
1964 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
1966 <div class="doc_text">
1968 <pre> <result> = urem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1971 <p>The '<tt>urem</tt>' instruction returns the remainder from the
1972 unsigned division of its two arguments.</p>
1974 <p>The two arguments to the '<tt>urem</tt>' instruction must be
1975 <a href="#t_integer">integer</a> values. Both arguments must have identical
1978 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
1979 This instruction always performs an unsigned division to get the remainder,
1980 regardless of whether the arguments are unsigned or not.</p>
1982 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
1986 <!-- _______________________________________________________________________ -->
1987 <div class="doc_subsubsection"> <a name="i_srem">'<tt>srem</tt>'
1988 Instruction</a> </div>
1989 <div class="doc_text">
1991 <pre> <result> = srem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1994 <p>The '<tt>srem</tt>' instruction returns the remainder from the
1995 signed division of its two operands.</p>
1997 <p>The two arguments to the '<tt>srem</tt>' instruction must be
1998 <a href="#t_integer">integer</a> values. Both arguments must have identical
2001 <p>This instruction returns the <i>remainder</i> of a division (where the result
2002 has the same sign as the dividend, <tt>var1</tt>), not the <i>modulo</i>
2003 operator (where the result has the same sign as the divisor, <tt>var2</tt>) of
2004 a value. For more information about the difference, see <a
2005 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2006 Math Forum</a>. For a table of how this is implemented in various languages,
2007 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2008 Wikipedia: modulo operation</a>.</p>
2010 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2014 <!-- _______________________________________________________________________ -->
2015 <div class="doc_subsubsection"> <a name="i_frem">'<tt>frem</tt>'
2016 Instruction</a> </div>
2017 <div class="doc_text">
2019 <pre> <result> = frem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2022 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2023 division of its two operands.</p>
2025 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2026 <a href="#t_floating">floating point</a> values. Both arguments must have
2027 identical types.</p>
2029 <p>This instruction returns the <i>remainder</i> of a division.</p>
2031 <pre> <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2035 <!-- ======================================================================= -->
2036 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2037 Operations</a> </div>
2038 <div class="doc_text">
2039 <p>Bitwise binary operators are used to do various forms of
2040 bit-twiddling in a program. They are generally very efficient
2041 instructions and can commonly be strength reduced from other
2042 instructions. They require two operands, execute an operation on them,
2043 and produce a single value. The resulting value of the bitwise binary
2044 operators is always the same type as its first operand.</p>
2047 <!-- _______________________________________________________________________ -->
2048 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2049 Instruction</a> </div>
2050 <div class="doc_text">
2052 <pre> <result> = shl <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2055 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2056 the left a specified number of bits.</p>
2058 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2059 href="#t_integer">integer</a> type.</p>
2061 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>.</p>
2062 <h5>Example:</h5><pre>
2063 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2064 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2065 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2068 <!-- _______________________________________________________________________ -->
2069 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2070 Instruction</a> </div>
2071 <div class="doc_text">
2073 <pre> <result> = lshr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2077 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2078 operand shifted to the right a specified number of bits.</p>
2081 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2082 <a href="#t_integer">integer</a> type.</p>
2085 <p>This instruction always performs a logical shift right operation. The most
2086 significant bits of the result will be filled with zero bits after the
2091 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2092 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2093 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2094 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2098 <!-- _______________________________________________________________________ -->
2099 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2100 Instruction</a> </div>
2101 <div class="doc_text">
2104 <pre> <result> = ashr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2108 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2109 operand shifted to the right a specified number of bits.</p>
2112 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2113 <a href="#t_integer">integer</a> type.</p>
2116 <p>This instruction always performs an arithmetic shift right operation,
2117 The most significant bits of the result will be filled with the sign bit
2118 of <tt>var1</tt>.</p>
2122 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2123 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2124 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2125 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2129 <!-- _______________________________________________________________________ -->
2130 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2131 Instruction</a> </div>
2132 <div class="doc_text">
2134 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2137 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2138 its two operands.</p>
2140 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
2141 href="#t_integer">integer</a> values. Both arguments must have
2142 identical types.</p>
2144 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2146 <div style="align: center">
2147 <table border="1" cellspacing="0" cellpadding="4">
2178 <pre> <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2179 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2180 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2183 <!-- _______________________________________________________________________ -->
2184 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2185 <div class="doc_text">
2187 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2190 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2191 or of its two operands.</p>
2193 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
2194 href="#t_integer">integer</a> values. Both arguments must have
2195 identical types.</p>
2197 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2199 <div style="align: center">
2200 <table border="1" cellspacing="0" cellpadding="4">
2231 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2232 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2233 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2236 <!-- _______________________________________________________________________ -->
2237 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2238 Instruction</a> </div>
2239 <div class="doc_text">
2241 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2244 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2245 or of its two operands. The <tt>xor</tt> is used to implement the
2246 "one's complement" operation, which is the "~" operator in C.</p>
2248 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
2249 href="#t_integer">integer</a> values. Both arguments must have
2250 identical types.</p>
2252 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2254 <div style="align: center">
2255 <table border="1" cellspacing="0" cellpadding="4">
2287 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2288 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2289 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2290 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2294 <!-- ======================================================================= -->
2295 <div class="doc_subsection">
2296 <a name="vectorops">Vector Operations</a>
2299 <div class="doc_text">
2301 <p>LLVM supports several instructions to represent vector operations in a
2302 target-independent manner. This instructions cover the element-access and
2303 vector-specific operations needed to process vectors effectively. While LLVM
2304 does directly support these vector operations, many sophisticated algorithms
2305 will want to use target-specific intrinsics to take full advantage of a specific
2310 <!-- _______________________________________________________________________ -->
2311 <div class="doc_subsubsection">
2312 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2315 <div class="doc_text">
2320 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2326 The '<tt>extractelement</tt>' instruction extracts a single scalar
2327 element from a vector at a specified index.
2334 The first operand of an '<tt>extractelement</tt>' instruction is a
2335 value of <a href="#t_vector">vector</a> type. The second operand is
2336 an index indicating the position from which to extract the element.
2337 The index may be a variable.</p>
2342 The result is a scalar of the same type as the element type of
2343 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2344 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2345 results are undefined.
2351 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
2356 <!-- _______________________________________________________________________ -->
2357 <div class="doc_subsubsection">
2358 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2361 <div class="doc_text">
2366 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2372 The '<tt>insertelement</tt>' instruction inserts a scalar
2373 element into a vector at a specified index.
2380 The first operand of an '<tt>insertelement</tt>' instruction is a
2381 value of <a href="#t_vector">vector</a> type. The second operand is a
2382 scalar value whose type must equal the element type of the first
2383 operand. The third operand is an index indicating the position at
2384 which to insert the value. The index may be a variable.</p>
2389 The result is a vector of the same type as <tt>val</tt>. Its
2390 element values are those of <tt>val</tt> except at position
2391 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2392 exceeds the length of <tt>val</tt>, the results are undefined.
2398 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
2402 <!-- _______________________________________________________________________ -->
2403 <div class="doc_subsubsection">
2404 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2407 <div class="doc_text">
2412 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x i32> <mask> <i>; yields <n x <ty>></i>
2418 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2419 from two input vectors, returning a vector of the same type.
2425 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2426 with types that match each other and types that match the result of the
2427 instruction. The third argument is a shuffle mask, which has the same number
2428 of elements as the other vector type, but whose element type is always 'i32'.
2432 The shuffle mask operand is required to be a constant vector with either
2433 constant integer or undef values.
2439 The elements of the two input vectors are numbered from left to right across
2440 both of the vectors. The shuffle mask operand specifies, for each element of
2441 the result vector, which element of the two input registers the result element
2442 gets. The element selector may be undef (meaning "don't care") and the second
2443 operand may be undef if performing a shuffle from only one vector.
2449 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
2450 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
2451 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
2452 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2457 <!-- ======================================================================= -->
2458 <div class="doc_subsection">
2459 <a name="memoryops">Memory Access and Addressing Operations</a>
2462 <div class="doc_text">
2464 <p>A key design point of an SSA-based representation is how it
2465 represents memory. In LLVM, no memory locations are in SSA form, which
2466 makes things very simple. This section describes how to read, write,
2467 allocate, and free memory in LLVM.</p>
2471 <!-- _______________________________________________________________________ -->
2472 <div class="doc_subsubsection">
2473 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
2476 <div class="doc_text">
2481 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2486 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
2487 heap and returns a pointer to it.</p>
2491 <p>The '<tt>malloc</tt>' instruction allocates
2492 <tt>sizeof(<type>)*NumElements</tt>
2493 bytes of memory from the operating system and returns a pointer of the
2494 appropriate type to the program. If "NumElements" is specified, it is the
2495 number of elements allocated. If an alignment is specified, the value result
2496 of the allocation is guaranteed to be aligned to at least that boundary. If
2497 not specified, or if zero, the target can choose to align the allocation on any
2498 convenient boundary.</p>
2500 <p>'<tt>type</tt>' must be a sized type.</p>
2504 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
2505 a pointer is returned.</p>
2510 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
2512 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
2513 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
2514 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
2515 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
2516 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
2520 <!-- _______________________________________________________________________ -->
2521 <div class="doc_subsubsection">
2522 <a name="i_free">'<tt>free</tt>' Instruction</a>
2525 <div class="doc_text">
2530 free <type> <value> <i>; yields {void}</i>
2535 <p>The '<tt>free</tt>' instruction returns memory back to the unused
2536 memory heap to be reallocated in the future.</p>
2540 <p>'<tt>value</tt>' shall be a pointer value that points to a value
2541 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
2546 <p>Access to the memory pointed to by the pointer is no longer defined
2547 after this instruction executes.</p>
2552 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
2553 free [4 x i8]* %array
2557 <!-- _______________________________________________________________________ -->
2558 <div class="doc_subsubsection">
2559 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
2562 <div class="doc_text">
2567 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2572 <p>The '<tt>alloca</tt>' instruction allocates memory on the current
2573 stack frame of the procedure that is live until the current function
2574 returns to its caller.</p>
2578 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
2579 bytes of memory on the runtime stack, returning a pointer of the
2580 appropriate type to the program. If "NumElements" is specified, it is the
2581 number of elements allocated. If an alignment is specified, the value result
2582 of the allocation is guaranteed to be aligned to at least that boundary. If
2583 not specified, or if zero, the target can choose to align the allocation on any
2584 convenient boundary.</p>
2586 <p>'<tt>type</tt>' may be any sized type.</p>
2590 <p>Memory is allocated; a pointer is returned. '<tt>alloca</tt>'d
2591 memory is automatically released when the function returns. The '<tt>alloca</tt>'
2592 instruction is commonly used to represent automatic variables that must
2593 have an address available. When the function returns (either with the <tt><a
2594 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
2595 instructions), the memory is reclaimed.</p>
2600 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
2601 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
2602 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
2603 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
2607 <!-- _______________________________________________________________________ -->
2608 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
2609 Instruction</a> </div>
2610 <div class="doc_text">
2612 <pre> <result> = load <ty>* <pointer><br> <result> = volatile load <ty>* <pointer><br></pre>
2614 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
2616 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
2617 address from which to load. The pointer must point to a <a
2618 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
2619 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
2620 the number or order of execution of this <tt>load</tt> with other
2621 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
2624 <p>The location of memory pointed to is loaded.</p>
2626 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2628 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
2629 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
2632 <!-- _______________________________________________________________________ -->
2633 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
2634 Instruction</a> </div>
2635 <div class="doc_text">
2637 <pre> store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
2638 volatile store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
2641 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
2643 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
2644 to store and an address in which to store it. The type of the '<tt><pointer></tt>'
2645 operand must be a pointer to the type of the '<tt><value></tt>'
2646 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
2647 optimizer is not allowed to modify the number or order of execution of
2648 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
2649 href="#i_store">store</a></tt> instructions.</p>
2651 <p>The contents of memory are updated to contain '<tt><value></tt>'
2652 at the location specified by the '<tt><pointer></tt>' operand.</p>
2654 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2656 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
2657 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
2661 <!-- _______________________________________________________________________ -->
2662 <div class="doc_subsubsection">
2663 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
2666 <div class="doc_text">
2669 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
2675 The '<tt>getelementptr</tt>' instruction is used to get the address of a
2676 subelement of an aggregate data structure.</p>
2680 <p>This instruction takes a list of integer operands that indicate what
2681 elements of the aggregate object to index to. The actual types of the arguments
2682 provided depend on the type of the first pointer argument. The
2683 '<tt>getelementptr</tt>' instruction is used to index down through the type
2684 levels of a structure or to a specific index in an array. When indexing into a
2685 structure, only <tt>i32</tt> integer constants are allowed. When indexing
2686 into an array or pointer, only integers of 32 or 64 bits are allowed, and will
2687 be sign extended to 64-bit values.</p>
2689 <p>For example, let's consider a C code fragment and how it gets
2690 compiled to LLVM:</p>
2704 define i32 *foo(struct ST *s) {
2705 return &s[1].Z.B[5][13];
2709 <p>The LLVM code generated by the GCC frontend is:</p>
2712 %RT = type { i8 , [10 x [20 x i32]], i8 }
2713 %ST = type { i32, double, %RT }
2715 define i32* %foo(%ST* %s) {
2717 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
2724 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
2725 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
2726 and <a href="#t_array">array</a> types can use a 32-bit or 64-bit
2727 <a href="#t_integer">integer</a> type but the value will always be sign extended
2728 to 64-bits. <a href="#t_struct">Structure</a> types, require <tt>i32</tt>
2729 <b>constants</b>.</p>
2731 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
2732 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
2733 }</tt>' type, a structure. The second index indexes into the third element of
2734 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
2735 i8 }</tt>' type, another structure. The third index indexes into the second
2736 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
2737 array. The two dimensions of the array are subscripted into, yielding an
2738 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
2739 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
2741 <p>Note that it is perfectly legal to index partially through a
2742 structure, returning a pointer to an inner element. Because of this,
2743 the LLVM code for the given testcase is equivalent to:</p>
2746 define i32* %foo(%ST* %s) {
2747 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
2748 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
2749 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
2750 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
2751 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
2756 <p>Note that it is undefined to access an array out of bounds: array and
2757 pointer indexes must always be within the defined bounds of the array type.
2758 The one exception for this rules is zero length arrays. These arrays are
2759 defined to be accessible as variable length arrays, which requires access
2760 beyond the zero'th element.</p>
2762 <p>The getelementptr instruction is often confusing. For some more insight
2763 into how it works, see <a href="GetElementPtr.html">the getelementptr
2769 <i>; yields [12 x i8]*:aptr</i>
2770 %aptr = getelementptr {i32, [12 x i8]}* %sptr, i64 0, i32 1
2774 <!-- ======================================================================= -->
2775 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
2777 <div class="doc_text">
2778 <p>The instructions in this category are the conversion instructions (casting)
2779 which all take a single operand and a type. They perform various bit conversions
2783 <!-- _______________________________________________________________________ -->
2784 <div class="doc_subsubsection">
2785 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
2787 <div class="doc_text">
2791 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
2796 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
2801 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
2802 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
2803 and type of the result, which must be an <a href="#t_integer">integer</a>
2804 type. The bit size of <tt>value</tt> must be larger than the bit size of
2805 <tt>ty2</tt>. Equal sized types are not allowed.</p>
2809 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
2810 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
2811 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
2812 It will always truncate bits.</p>
2816 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
2817 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
2818 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
2822 <!-- _______________________________________________________________________ -->
2823 <div class="doc_subsubsection">
2824 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
2826 <div class="doc_text">
2830 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
2834 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
2839 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
2840 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
2841 also be of <a href="#t_integer">integer</a> type. The bit size of the
2842 <tt>value</tt> must be smaller than the bit size of the destination type,
2846 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
2847 bits until it reaches the size of the destination type, <tt>ty2</tt>. When the
2848 the operand and the type are the same size, no bit filling is done and the
2849 cast is considered a <i>no-op cast</i> because no bits change (only the type
2852 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
2856 %X = zext i32 257 to i64 <i>; yields i64:257</i>
2857 %Y = zext i1 true to i32 <i>; yields i32:1</i>
2861 <!-- _______________________________________________________________________ -->
2862 <div class="doc_subsubsection">
2863 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
2865 <div class="doc_text">
2869 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
2873 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
2877 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
2878 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
2879 also be of <a href="#t_integer">integer</a> type. The bit size of the
2880 <tt>value</tt> must be smaller than the bit size of the destination type,
2885 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
2886 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
2887 the type <tt>ty2</tt>. When the the operand and the type are the same size,
2888 no bit filling is done and the cast is considered a <i>no-op cast</i> because
2889 no bits change (only the type changes).</p>
2891 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
2895 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
2896 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
2900 <!-- _______________________________________________________________________ -->
2901 <div class="doc_subsubsection">
2902 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
2905 <div class="doc_text">
2910 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
2914 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
2919 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
2920 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
2921 cast it to. The size of <tt>value</tt> must be larger than the size of
2922 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
2923 <i>no-op cast</i>.</p>
2926 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
2927 <a href="#t_floating">floating point</a> type to a smaller
2928 <a href="#t_floating">floating point</a> type. If the value cannot fit within
2929 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
2933 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
2934 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
2938 <!-- _______________________________________________________________________ -->
2939 <div class="doc_subsubsection">
2940 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
2942 <div class="doc_text">
2946 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
2950 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
2951 floating point value.</p>
2954 <p>The '<tt>fpext</tt>' instruction takes a
2955 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
2956 and a <a href="#t_floating">floating point</a> type to cast it to. The source
2957 type must be smaller than the destination type.</p>
2960 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
2961 <a href="#t_floating">floating point</a> type to a larger
2962 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
2963 used to make a <i>no-op cast</i> because it always changes bits. Use
2964 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
2968 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
2969 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
2973 <!-- _______________________________________________________________________ -->
2974 <div class="doc_subsubsection">
2975 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
2977 <div class="doc_text">
2981 <result> = fp2uint <ty> <value> to <ty2> <i>; yields ty2</i>
2985 <p>The '<tt>fp2uint</tt>' converts a floating point <tt>value</tt> to its
2986 unsigned integer equivalent of type <tt>ty2</tt>.
2990 <p>The '<tt>fp2uint</tt>' instruction takes a value to cast, which must be a
2991 <a href="#t_floating">floating point</a> value, and a type to cast it to, which
2992 must be an <a href="#t_integer">integer</a> type.</p>
2995 <p> The '<tt>fp2uint</tt>' instruction converts its
2996 <a href="#t_floating">floating point</a> operand into the nearest (rounding
2997 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
2998 the results are undefined.</p>
3000 <p>When converting to i1, the conversion is done as a comparison against
3001 zero. If the <tt>value</tt> was zero, the i1 result will be <tt>false</tt>.
3002 If the <tt>value</tt> was non-zero, the i1 result will be <tt>true</tt>.</p>
3006 %X = fp2uint double 123.0 to i32 <i>; yields i32:123</i>
3007 %Y = fp2uint float 1.0E+300 to i1 <i>; yields i1:true</i>
3008 %X = fp2uint float 1.04E+17 to i8 <i>; yields undefined:1</i>
3012 <!-- _______________________________________________________________________ -->
3013 <div class="doc_subsubsection">
3014 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3016 <div class="doc_text">
3020 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3024 <p>The '<tt>fptosi</tt>' instruction converts
3025 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3030 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3031 <a href="#t_floating">floating point</a> value, and a type to cast it to, which
3032 must also be an <a href="#t_integer">integer</a> type.</p>
3035 <p>The '<tt>fptosi</tt>' instruction converts its
3036 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3037 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3038 the results are undefined.</p>
3040 <p>When converting to i1, the conversion is done as a comparison against
3041 zero. If the <tt>value</tt> was zero, the i1 result will be <tt>false</tt>.
3042 If the <tt>value</tt> was non-zero, the i1 result will be <tt>true</tt>.</p>
3046 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3047 %Y = fptosi float 1.0E-247 to i1 <i>; yields i1:true</i>
3048 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3052 <!-- _______________________________________________________________________ -->
3053 <div class="doc_subsubsection">
3054 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3056 <div class="doc_text">
3060 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3064 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3065 integer and converts that value to the <tt>ty2</tt> type.</p>
3069 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be an
3070 <a href="#t_integer">integer</a> value, and a type to cast it to, which must
3071 be a <a href="#t_floating">floating point</a> type.</p>
3074 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3075 integer quantity and converts it to the corresponding floating point value. If
3076 the value cannot fit in the floating point value, the results are undefined.</p>
3081 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3082 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3086 <!-- _______________________________________________________________________ -->
3087 <div class="doc_subsubsection">
3088 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3090 <div class="doc_text">
3094 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3098 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3099 integer and converts that value to the <tt>ty2</tt> type.</p>
3102 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be an
3103 <a href="#t_integer">integer</a> value, and a type to cast it to, which must be
3104 a <a href="#t_floating">floating point</a> type.</p>
3107 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3108 integer quantity and converts it to the corresponding floating point value. If
3109 the value cannot fit in the floating point value, the results are undefined.</p>
3113 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3114 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3118 <!-- _______________________________________________________________________ -->
3119 <div class="doc_subsubsection">
3120 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3122 <div class="doc_text">
3126 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3130 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3131 the integer type <tt>ty2</tt>.</p>
3134 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3135 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3136 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
3139 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3140 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3141 truncating or zero extending that value to the size of the integer type. If
3142 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3143 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3144 are the same size, then nothing is done (<i>no-op cast</i>).</p>
3148 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit</i>
3149 %Y = ptrtoint i32* %x to i64 <i>; yields zero extend on 32-bit</i>
3153 <!-- _______________________________________________________________________ -->
3154 <div class="doc_subsubsection">
3155 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3157 <div class="doc_text">
3161 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3165 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3166 a pointer type, <tt>ty2</tt>.</p>
3169 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3170 value to cast, and a type to cast it to, which must be a
3171 <a href="#t_pointer">pointer</a> type.
3174 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3175 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3176 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3177 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3178 the size of a pointer then a zero extension is done. If they are the same size,
3179 nothing is done (<i>no-op cast</i>).</p>
3183 %X = inttoptr i32 255 to i32* <i>; yields zero extend on 64-bit</i>
3184 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit </i>
3185 %Y = inttoptr i16 0 to i32* <i>; yields zero extend on 32-bit</i>
3189 <!-- _______________________________________________________________________ -->
3190 <div class="doc_subsubsection">
3191 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3193 <div class="doc_text">
3197 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
3201 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3202 <tt>ty2</tt> without changing any bits.</p>
3205 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3206 a first class value, and a type to cast it to, which must also be a <a
3207 href="#t_firstclass">first class</a> type. The bit sizes of <tt>value</tt>
3208 and the destination type, <tt>ty2</tt>, must be identical. If the source
3209 type is a pointer, the destination type must also be a pointer.</p>
3212 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3213 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3214 this conversion. The conversion is done as if the <tt>value</tt> had been
3215 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3216 converted to other pointer types with this instruction. To convert pointers to
3217 other types, use the <a href="#i_inttoptr">inttoptr</a> or
3218 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3222 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3223 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3224 %Z = bitcast <2xint> %V to i64; <i>; yields i64: %V</i>
3228 <!-- ======================================================================= -->
3229 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3230 <div class="doc_text">
3231 <p>The instructions in this category are the "miscellaneous"
3232 instructions, which defy better classification.</p>
3235 <!-- _______________________________________________________________________ -->
3236 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3238 <div class="doc_text">
3240 <pre> <result> = icmp <cond> <ty> <var1>, <var2>
3241 <i>; yields {i1}:result</i>
3244 <p>The '<tt>icmp</tt>' instruction returns a boolean value based on comparison
3245 of its two integer operands.</p>
3247 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3248 the condition code which indicates the kind of comparison to perform. It is not
3249 a value, just a keyword. The possibilities for the condition code are:
3251 <li><tt>eq</tt>: equal</li>
3252 <li><tt>ne</tt>: not equal </li>
3253 <li><tt>ugt</tt>: unsigned greater than</li>
3254 <li><tt>uge</tt>: unsigned greater or equal</li>
3255 <li><tt>ult</tt>: unsigned less than</li>
3256 <li><tt>ule</tt>: unsigned less or equal</li>
3257 <li><tt>sgt</tt>: signed greater than</li>
3258 <li><tt>sge</tt>: signed greater or equal</li>
3259 <li><tt>slt</tt>: signed less than</li>
3260 <li><tt>sle</tt>: signed less or equal</li>
3262 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3263 <a href="#t_pointer">pointer</a> typed. They must also be identical types.</p>
3265 <p>The '<tt>icmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3266 the condition code given as <tt>cond</tt>. The comparison performed always
3267 yields a <a href="#t_primitive">i1</a> result, as follows:
3269 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3270 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3272 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3273 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3274 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3275 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3276 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3277 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3278 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3279 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3280 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3281 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3282 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3283 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3284 <li><tt>sge</tt>: interprets the operands as signed values and yields
3285 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3286 <li><tt>slt</tt>: interprets the operands as signed values and yields
3287 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3288 <li><tt>sle</tt>: interprets the operands as signed values and yields
3289 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3291 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3292 values are treated as integers and then compared.</p>
3295 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
3296 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
3297 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
3298 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3299 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3300 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
3304 <!-- _______________________________________________________________________ -->
3305 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3307 <div class="doc_text">
3309 <pre> <result> = fcmp <cond> <ty> <var1>, <var2>
3310 <i>; yields {i1}:result</i>
3313 <p>The '<tt>fcmp</tt>' instruction returns a boolean value based on comparison
3314 of its floating point operands.</p>
3316 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
3317 the condition code which indicates the kind of comparison to perform. It is not
3318 a value, just a keyword. The possibilities for the condition code are:
3320 <li><tt>false</tt>: no comparison, always returns false</li>
3321 <li><tt>oeq</tt>: ordered and equal</li>
3322 <li><tt>ogt</tt>: ordered and greater than </li>
3323 <li><tt>oge</tt>: ordered and greater than or equal</li>
3324 <li><tt>olt</tt>: ordered and less than </li>
3325 <li><tt>ole</tt>: ordered and less than or equal</li>
3326 <li><tt>one</tt>: ordered and not equal</li>
3327 <li><tt>ord</tt>: ordered (no nans)</li>
3328 <li><tt>ueq</tt>: unordered or equal</li>
3329 <li><tt>ugt</tt>: unordered or greater than </li>
3330 <li><tt>uge</tt>: unordered or greater than or equal</li>
3331 <li><tt>ult</tt>: unordered or less than </li>
3332 <li><tt>ule</tt>: unordered or less than or equal</li>
3333 <li><tt>une</tt>: unordered or not equal</li>
3334 <li><tt>uno</tt>: unordered (either nans)</li>
3335 <li><tt>true</tt>: no comparison, always returns true</li>
3337 <p>In the preceding, <i>ordered</i> means that neither operand is a QNAN while
3338 <i>unordered</i> means that either operand may be a QNAN.</p>
3339 <p>The <tt>val1</tt> and <tt>val2</tt> arguments must be
3340 <a href="#t_floating">floating point</a> typed. They must have identical
3342 <p>In the foregoing, <i>ordered</i> means that neither operand is a QNAN and
3343 <i>unordered</i> means that either operand is a QNAN.</p>
3345 <p>The '<tt>fcmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3346 the condition code given as <tt>cond</tt>. The comparison performed always
3347 yields a <a href="#t_primitive">i1</a> result, as follows:
3349 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
3350 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3351 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3352 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3353 <tt>var1</tt> is greather than <tt>var2</tt>.</li>
3354 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3355 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3356 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3357 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3358 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3359 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3360 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3361 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3362 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
3363 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
3364 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3365 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
3366 <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3367 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
3368 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3369 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
3370 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3371 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
3372 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3373 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
3374 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3375 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
3376 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
3380 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
3381 <result> = icmp one float 4.0, 5.0 <i>; yields: result=true</i>
3382 <result> = icmp olt float 4.0, 5.0 <i>; yields: result=true</i>
3383 <result> = icmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
3387 <!-- _______________________________________________________________________ -->
3388 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
3389 Instruction</a> </div>
3390 <div class="doc_text">
3392 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
3394 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
3395 the SSA graph representing the function.</p>
3397 <p>The type of the incoming values are specified with the first type
3398 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
3399 as arguments, with one pair for each predecessor basic block of the
3400 current block. Only values of <a href="#t_firstclass">first class</a>
3401 type may be used as the value arguments to the PHI node. Only labels
3402 may be used as the label arguments.</p>
3403 <p>There must be no non-phi instructions between the start of a basic
3404 block and the PHI instructions: i.e. PHI instructions must be first in
3407 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the
3408 value specified by the parameter, depending on which basic block we
3409 came from in the last <a href="#terminators">terminator</a> instruction.</p>
3411 <pre>Loop: ; Infinite loop that counts from 0 on up...<br> %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]<br> %nextindvar = add i32 %indvar, 1<br> br label %Loop<br></pre>
3414 <!-- _______________________________________________________________________ -->
3415 <div class="doc_subsubsection">
3416 <a name="i_select">'<tt>select</tt>' Instruction</a>
3419 <div class="doc_text">
3424 <result> = select i1 <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
3430 The '<tt>select</tt>' instruction is used to choose one value based on a
3431 condition, without branching.
3438 The '<tt>select</tt>' instruction requires a boolean value indicating the condition, and two values of the same <a href="#t_firstclass">first class</a> type.
3444 If the boolean condition evaluates to true, the instruction returns the first
3445 value argument; otherwise, it returns the second value argument.
3451 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
3456 <!-- _______________________________________________________________________ -->
3457 <div class="doc_subsubsection">
3458 <a name="i_call">'<tt>call</tt>' Instruction</a>
3461 <div class="doc_text">
3465 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty>* <fnptrval>(<param list>)
3470 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
3474 <p>This instruction requires several arguments:</p>
3478 <p>The optional "tail" marker indicates whether the callee function accesses
3479 any allocas or varargs in the caller. If the "tail" marker is present, the
3480 function call is eligible for tail call optimization. Note that calls may
3481 be marked "tail" even if they do not occur before a <a
3482 href="#i_ret"><tt>ret</tt></a> instruction.
3485 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
3486 convention</a> the call should use. If none is specified, the call defaults
3487 to using C calling conventions.
3490 <p>'<tt>ty</tt>': shall be the signature of the pointer to function value
3491 being invoked. The argument types must match the types implied by this
3492 signature. This type can be omitted if the function is not varargs and
3493 if the function type does not return a pointer to a function.</p>
3496 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
3497 be invoked. In most cases, this is a direct function invocation, but
3498 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
3499 to function value.</p>
3502 <p>'<tt>function args</tt>': argument list whose types match the
3503 function signature argument types. All arguments must be of
3504 <a href="#t_firstclass">first class</a> type. If the function signature
3505 indicates the function accepts a variable number of arguments, the extra
3506 arguments can be specified.</p>
3512 <p>The '<tt>call</tt>' instruction is used to cause control flow to
3513 transfer to a specified function, with its incoming arguments bound to
3514 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
3515 instruction in the called function, control flow continues with the
3516 instruction after the function call, and the return value of the
3517 function is bound to the result argument. This is a simpler case of
3518 the <a href="#i_invoke">invoke</a> instruction.</p>
3523 %retval = call i32 %test(i32 %argc)
3524 call i32(i8 *, ...) *%printf(i8 * %msg, i32 12, i8 42);
3525 %X = tail call i32 %foo()
3526 %Y = tail call <a href="#callingconv">fastcc</a> i32 %foo()
3531 <!-- _______________________________________________________________________ -->
3532 <div class="doc_subsubsection">
3533 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
3536 <div class="doc_text">
3541 <resultval> = va_arg <va_list*> <arglist>, <argty>
3546 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
3547 the "variable argument" area of a function call. It is used to implement the
3548 <tt>va_arg</tt> macro in C.</p>
3552 <p>This instruction takes a <tt>va_list*</tt> value and the type of
3553 the argument. It returns a value of the specified argument type and
3554 increments the <tt>va_list</tt> to point to the next argument. Again, the
3555 actual type of <tt>va_list</tt> is target specific.</p>
3559 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
3560 type from the specified <tt>va_list</tt> and causes the
3561 <tt>va_list</tt> to point to the next argument. For more information,
3562 see the variable argument handling <a href="#int_varargs">Intrinsic
3565 <p>It is legal for this instruction to be called in a function which does not
3566 take a variable number of arguments, for example, the <tt>vfprintf</tt>
3569 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
3570 href="#intrinsics">intrinsic function</a> because it takes a type as an
3575 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
3579 <!-- *********************************************************************** -->
3580 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
3581 <!-- *********************************************************************** -->
3583 <div class="doc_text">
3585 <p>LLVM supports the notion of an "intrinsic function". These functions have
3586 well known names and semantics and are required to follow certain restrictions.
3587 Overall, these intrinsics represent an extension mechanism for the LLVM
3588 language that does not require changing all of the transformations in LLVM to
3589 add to the language (or the bytecode reader/writer, the parser,
3592 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
3593 prefix is reserved in LLVM for intrinsic names; thus, functions may not be named
3594 this. Intrinsic functions must always be external functions: you cannot define
3595 the body of intrinsic functions. Intrinsic functions may only be used in call
3596 or invoke instructions: it is illegal to take the address of an intrinsic
3597 function. Additionally, because intrinsic functions are part of the LLVM
3598 language, it is required that they all be documented here if any are added.</p>
3600 <p>Some intrinsic functions can be overloaded. That is, the intrinsic represents
3601 a family of functions that perform the same operation but on different data
3602 types. This is most frequent with the integer types. Since LLVM can represent
3603 over 8 million different integer types, there is a way to declare an intrinsic
3604 that can be overloaded based on its arguments. Such intrinsics will have the
3605 names of the arbitrary types encoded into the intrinsic function name, each
3606 preceded by a period. For example, the <tt>llvm.ctpop</tt> function can take an
3607 integer of any width. This leads to a family of functions such as
3608 <tt>i32 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i32 @llvm.ctpop.i29(i29 %val)</tt>.
3612 <p>To learn how to add an intrinsic function, please see the
3613 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
3618 <!-- ======================================================================= -->
3619 <div class="doc_subsection">
3620 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
3623 <div class="doc_text">
3625 <p>Variable argument support is defined in LLVM with the <a
3626 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
3627 intrinsic functions. These functions are related to the similarly
3628 named macros defined in the <tt><stdarg.h></tt> header file.</p>
3630 <p>All of these functions operate on arguments that use a
3631 target-specific value type "<tt>va_list</tt>". The LLVM assembly
3632 language reference manual does not define what this type is, so all
3633 transformations should be prepared to handle intrinsics with any type
3636 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
3637 instruction and the variable argument handling intrinsic functions are
3641 define i32 @test(i32 %X, ...) {
3642 ; Initialize variable argument processing
3644 %ap2 = bitcast i8** %ap to i8*
3645 call void @llvm.va_start(i8* %ap2)
3647 ; Read a single integer argument
3648 %tmp = va_arg i8 ** %ap, i32
3650 ; Demonstrate usage of llvm.va_copy and llvm.va_end
3652 %aq2 = bitcast i8** %aq to i8*
3653 call void @llvm.va_copy(i8 *%aq2, i8* %ap2)
3654 call void @llvm.va_end(i8* %aq2)
3656 ; Stop processing of arguments.
3657 call void @llvm.va_end(i8* %ap2)
3661 declare void @llvm.va_start(i8*)
3662 declare void @llvm.va_copy(i8*, i8*)
3663 declare void @llvm.va_end(i8*)
3667 <!-- _______________________________________________________________________ -->
3668 <div class="doc_subsubsection">
3669 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
3673 <div class="doc_text">
3675 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
3677 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
3678 <tt>*<arglist></tt> for subsequent use by <tt><a
3679 href="#i_va_arg">va_arg</a></tt>.</p>
3683 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
3687 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
3688 macro available in C. In a target-dependent way, it initializes the
3689 <tt>va_list</tt> element the argument points to, so that the next call to
3690 <tt>va_arg</tt> will produce the first variable argument passed to the function.
3691 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
3692 last argument of the function, the compiler can figure that out.</p>
3696 <!-- _______________________________________________________________________ -->
3697 <div class="doc_subsubsection">
3698 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
3701 <div class="doc_text">
3703 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
3706 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt><arglist></tt>
3707 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
3708 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
3712 <p>The argument is a <tt>va_list</tt> to destroy.</p>
3716 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
3717 macro available in C. In a target-dependent way, it destroys the <tt>va_list</tt>.
3718 Calls to <a href="#int_va_start"><tt>llvm.va_start</tt></a> and <a
3719 href="#int_va_copy"><tt>llvm.va_copy</tt></a> must be matched exactly
3720 with calls to <tt>llvm.va_end</tt>.</p>
3724 <!-- _______________________________________________________________________ -->
3725 <div class="doc_subsubsection">
3726 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
3729 <div class="doc_text">
3734 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
3739 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position from
3740 the source argument list to the destination argument list.</p>
3744 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
3745 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
3750 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt> macro
3751 available in C. In a target-dependent way, it copies the source
3752 <tt>va_list</tt> element into the destination list. This intrinsic is necessary
3753 because the <tt><a href="#int_va_start">llvm.va_start</a></tt> intrinsic may be
3754 arbitrarily complex and require memory allocation, for example.</p>
3758 <!-- ======================================================================= -->
3759 <div class="doc_subsection">
3760 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
3763 <div class="doc_text">
3766 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
3767 Collection</a> requires the implementation and generation of these intrinsics.
3768 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
3769 stack</a>, as well as garbage collector implementations that require <a
3770 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
3771 Front-ends for type-safe garbage collected languages should generate these
3772 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
3773 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
3777 <!-- _______________________________________________________________________ -->
3778 <div class="doc_subsubsection">
3779 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
3782 <div class="doc_text">
3787 declare void @llvm.gcroot(<ty>** %ptrloc, <ty2>* %metadata)
3792 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
3793 the code generator, and allows some metadata to be associated with it.</p>
3797 <p>The first argument specifies the address of a stack object that contains the
3798 root pointer. The second pointer (which must be either a constant or a global
3799 value address) contains the meta-data to be associated with the root.</p>
3803 <p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
3804 location. At compile-time, the code generator generates information to allow
3805 the runtime to find the pointer at GC safe points.
3811 <!-- _______________________________________________________________________ -->
3812 <div class="doc_subsubsection">
3813 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
3816 <div class="doc_text">
3821 declare i8 * @llvm.gcread(i8 * %ObjPtr, i8 ** %Ptr)
3826 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
3827 locations, allowing garbage collector implementations that require read
3832 <p>The second argument is the address to read from, which should be an address
3833 allocated from the garbage collector. The first object is a pointer to the
3834 start of the referenced object, if needed by the language runtime (otherwise
3839 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
3840 instruction, but may be replaced with substantially more complex code by the
3841 garbage collector runtime, as needed.</p>
3846 <!-- _______________________________________________________________________ -->
3847 <div class="doc_subsubsection">
3848 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
3851 <div class="doc_text">
3856 declare void @llvm.gcwrite(i8 * %P1, i8 * %Obj, i8 ** %P2)
3861 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
3862 locations, allowing garbage collector implementations that require write
3863 barriers (such as generational or reference counting collectors).</p>
3867 <p>The first argument is the reference to store, the second is the start of the
3868 object to store it to, and the third is the address of the field of Obj to
3869 store to. If the runtime does not require a pointer to the object, Obj may be
3874 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
3875 instruction, but may be replaced with substantially more complex code by the
3876 garbage collector runtime, as needed.</p>
3882 <!-- ======================================================================= -->
3883 <div class="doc_subsection">
3884 <a name="int_codegen">Code Generator Intrinsics</a>
3887 <div class="doc_text">
3889 These intrinsics are provided by LLVM to expose special features that may only
3890 be implemented with code generator support.
3895 <!-- _______________________________________________________________________ -->
3896 <div class="doc_subsubsection">
3897 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
3900 <div class="doc_text">
3904 declare i8 *@llvm.returnaddress(i32 <level>)
3910 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
3911 target-specific value indicating the return address of the current function
3912 or one of its callers.
3918 The argument to this intrinsic indicates which function to return the address
3919 for. Zero indicates the calling function, one indicates its caller, etc. The
3920 argument is <b>required</b> to be a constant integer value.
3926 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
3927 the return address of the specified call frame, or zero if it cannot be
3928 identified. The value returned by this intrinsic is likely to be incorrect or 0
3929 for arguments other than zero, so it should only be used for debugging purposes.
3933 Note that calling this intrinsic does not prevent function inlining or other
3934 aggressive transformations, so the value returned may not be that of the obvious
3935 source-language caller.
3940 <!-- _______________________________________________________________________ -->
3941 <div class="doc_subsubsection">
3942 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
3945 <div class="doc_text">
3949 declare i8 *@llvm.frameaddress(i32 <level>)
3955 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
3956 target-specific frame pointer value for the specified stack frame.
3962 The argument to this intrinsic indicates which function to return the frame
3963 pointer for. Zero indicates the calling function, one indicates its caller,
3964 etc. The argument is <b>required</b> to be a constant integer value.
3970 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
3971 the frame address of the specified call frame, or zero if it cannot be
3972 identified. The value returned by this intrinsic is likely to be incorrect or 0
3973 for arguments other than zero, so it should only be used for debugging purposes.
3977 Note that calling this intrinsic does not prevent function inlining or other
3978 aggressive transformations, so the value returned may not be that of the obvious
3979 source-language caller.
3983 <!-- _______________________________________________________________________ -->
3984 <div class="doc_subsubsection">
3985 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
3988 <div class="doc_text">
3992 declare i8 *@llvm.stacksave()
3998 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
3999 the function stack, for use with <a href="#int_stackrestore">
4000 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4001 features like scoped automatic variable sized arrays in C99.
4007 This intrinsic returns a opaque pointer value that can be passed to <a
4008 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4009 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4010 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4011 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4012 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4013 that were allocated after the <tt>llvm.stacksave</tt> was executed.
4018 <!-- _______________________________________________________________________ -->
4019 <div class="doc_subsubsection">
4020 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4023 <div class="doc_text">
4027 declare void @llvm.stackrestore(i8 * %ptr)
4033 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4034 the function stack to the state it was in when the corresponding <a
4035 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4036 useful for implementing language features like scoped automatic variable sized
4043 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4049 <!-- _______________________________________________________________________ -->
4050 <div class="doc_subsubsection">
4051 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4054 <div class="doc_text">
4058 declare void @llvm.prefetch(i8 * <address>,
4059 i32 <rw>, i32 <locality>)
4066 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4067 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4069 effect on the behavior of the program but can change its performance
4076 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4077 determining if the fetch should be for a read (0) or write (1), and
4078 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4079 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4080 <tt>locality</tt> arguments must be constant integers.
4086 This intrinsic does not modify the behavior of the program. In particular,
4087 prefetches cannot trap and do not produce a value. On targets that support this
4088 intrinsic, the prefetch can provide hints to the processor cache for better
4094 <!-- _______________________________________________________________________ -->
4095 <div class="doc_subsubsection">
4096 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4099 <div class="doc_text">
4103 declare void @llvm.pcmarker( i32 <id> )
4110 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
4112 code to simulators and other tools. The method is target specific, but it is
4113 expected that the marker will use exported symbols to transmit the PC of the marker.
4114 The marker makes no guarantees that it will remain with any specific instruction
4115 after optimizations. It is possible that the presence of a marker will inhibit
4116 optimizations. The intended use is to be inserted after optimizations to allow
4117 correlations of simulation runs.
4123 <tt>id</tt> is a numerical id identifying the marker.
4129 This intrinsic does not modify the behavior of the program. Backends that do not
4130 support this intrinisic may ignore it.
4135 <!-- _______________________________________________________________________ -->
4136 <div class="doc_subsubsection">
4137 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
4140 <div class="doc_text">
4144 declare i64 @llvm.readcyclecounter( )
4151 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
4152 counter register (or similar low latency, high accuracy clocks) on those targets
4153 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4154 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4155 should only be used for small timings.
4161 When directly supported, reading the cycle counter should not modify any memory.
4162 Implementations are allowed to either return a application specific value or a
4163 system wide value. On backends without support, this is lowered to a constant 0.
4168 <!-- ======================================================================= -->
4169 <div class="doc_subsection">
4170 <a name="int_libc">Standard C Library Intrinsics</a>
4173 <div class="doc_text">
4175 LLVM provides intrinsics for a few important standard C library functions.
4176 These intrinsics allow source-language front-ends to pass information about the
4177 alignment of the pointer arguments to the code generator, providing opportunity
4178 for more efficient code generation.
4183 <!-- _______________________________________________________________________ -->
4184 <div class="doc_subsubsection">
4185 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
4188 <div class="doc_text">
4192 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
4193 i32 <len>, i32 <align>)
4194 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
4195 i64 <len>, i32 <align>)
4201 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4202 location to the destination location.
4206 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
4207 intrinsics do not return a value, and takes an extra alignment argument.
4213 The first argument is a pointer to the destination, the second is a pointer to
4214 the source. The third argument is an integer argument
4215 specifying the number of bytes to copy, and the fourth argument is the alignment
4216 of the source and destination locations.
4220 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4221 the caller guarantees that both the source and destination pointers are aligned
4228 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4229 location to the destination location, which are not allowed to overlap. It
4230 copies "len" bytes of memory over. If the argument is known to be aligned to
4231 some boundary, this can be specified as the fourth argument, otherwise it should
4237 <!-- _______________________________________________________________________ -->
4238 <div class="doc_subsubsection">
4239 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
4242 <div class="doc_text">
4246 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
4247 i32 <len>, i32 <align>)
4248 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
4249 i64 <len>, i32 <align>)
4255 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
4256 location to the destination location. It is similar to the
4257 '<tt>llvm.memcmp</tt>' intrinsic but allows the two memory locations to overlap.
4261 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
4262 intrinsics do not return a value, and takes an extra alignment argument.
4268 The first argument is a pointer to the destination, the second is a pointer to
4269 the source. The third argument is an integer argument
4270 specifying the number of bytes to copy, and the fourth argument is the alignment
4271 of the source and destination locations.
4275 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4276 the caller guarantees that the source and destination pointers are aligned to
4283 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
4284 location to the destination location, which may overlap. It
4285 copies "len" bytes of memory over. If the argument is known to be aligned to
4286 some boundary, this can be specified as the fourth argument, otherwise it should
4292 <!-- _______________________________________________________________________ -->
4293 <div class="doc_subsubsection">
4294 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
4297 <div class="doc_text">
4301 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
4302 i32 <len>, i32 <align>)
4303 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
4304 i64 <len>, i32 <align>)
4310 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
4315 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
4316 does not return a value, and takes an extra alignment argument.
4322 The first argument is a pointer to the destination to fill, the second is the
4323 byte value to fill it with, the third argument is an integer
4324 argument specifying the number of bytes to fill, and the fourth argument is the
4325 known alignment of destination location.
4329 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4330 the caller guarantees that the destination pointer is aligned to that boundary.
4336 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
4338 destination location. If the argument is known to be aligned to some boundary,
4339 this can be specified as the fourth argument, otherwise it should be set to 0 or
4345 <!-- _______________________________________________________________________ -->
4346 <div class="doc_subsubsection">
4347 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
4350 <div class="doc_text">
4354 declare float @llvm.sqrt.f32(float %Val)
4355 declare double @llvm.sqrt.f64(double %Val)
4361 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
4362 returning the same value as the libm '<tt>sqrt</tt>' function would. Unlike
4363 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
4364 negative numbers (which allows for better optimization).
4370 The argument and return value are floating point numbers of the same type.
4376 This function returns the sqrt of the specified operand if it is a positive
4377 floating point number.
4381 <!-- _______________________________________________________________________ -->
4382 <div class="doc_subsubsection">
4383 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
4386 <div class="doc_text">
4390 declare float @llvm.powi.f32(float %Val, i32 %power)
4391 declare double @llvm.powi.f64(double %Val, i32 %power)
4397 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
4398 specified (positive or negative) power. The order of evaluation of
4399 multiplications is not defined.
4405 The second argument is an integer power, and the first is a value to raise to
4412 This function returns the first value raised to the second power with an
4413 unspecified sequence of rounding operations.</p>
4417 <!-- ======================================================================= -->
4418 <div class="doc_subsection">
4419 <a name="int_manip">Bit Manipulation Intrinsics</a>
4422 <div class="doc_text">
4424 LLVM provides intrinsics for a few important bit manipulation operations.
4425 These allow efficient code generation for some algorithms.
4430 <!-- _______________________________________________________________________ -->
4431 <div class="doc_subsubsection">
4432 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
4435 <div class="doc_text">
4438 <p>This is an overloaded intrinsic function. You can use bswap on any integer
4439 type that is an even number of bytes (i.e. BitWidth % 16 == 0). Note the suffix
4440 that includes the type for the result and the operand.
4442 declare i16 @llvm.bswap.i16.i16(i16 <id>)
4443 declare i32 @llvm.bswap.i32.i32(i32 <id>)
4444 declare i64 @llvm.bswap.i64.i64(i64 <id>)
4450 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
4451 values with an even number of bytes (positive multiple of 16 bits). These are
4452 useful for performing operations on data that is not in the target's native
4459 The <tt>llvm.bswap.16.i16</tt> intrinsic returns an i16 value that has the high
4460 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
4461 intrinsic returns an i32 value that has the four bytes of the input i32
4462 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
4463 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48.i48</tt>,
4464 <tt>llvm.bswap.i64.i64</tt> and other intrinsics extend this concept to
4465 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
4470 <!-- _______________________________________________________________________ -->
4471 <div class="doc_subsubsection">
4472 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
4475 <div class="doc_text">
4478 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
4479 width. Not all targets support all bit widths however.
4481 declare i32 @llvm.ctpop.i8 (i8 <src>)
4482 declare i32 @llvm.ctpop.i16(i16 <src>)
4483 declare i32 @llvm.ctpop.i32(i32 <src>)
4484 declare i32 @llvm.ctpop.i64(i64 <src>)
4485 declare i32 @llvm.ctpop.i256(i256 <src>)
4491 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
4498 The only argument is the value to be counted. The argument may be of any
4499 integer type. The return type must match the argument type.
4505 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
4509 <!-- _______________________________________________________________________ -->
4510 <div class="doc_subsubsection">
4511 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
4514 <div class="doc_text">
4517 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
4518 integer bit width. Not all targets support all bit widths however.
4520 declare i32 @llvm.ctlz.i8 (i8 <src>)
4521 declare i32 @llvm.ctlz.i16(i16 <src>)
4522 declare i32 @llvm.ctlz.i32(i32 <src>)
4523 declare i32 @llvm.ctlz.i64(i64 <src>)
4524 declare i32 @llvm.ctlz.i256(i256 <src>)
4530 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
4531 leading zeros in a variable.
4537 The only argument is the value to be counted. The argument may be of any
4538 integer type. The return type must match the argument type.
4544 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
4545 in a variable. If the src == 0 then the result is the size in bits of the type
4546 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
4552 <!-- _______________________________________________________________________ -->
4553 <div class="doc_subsubsection">
4554 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
4557 <div class="doc_text">
4560 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
4561 integer bit width. Not all targets support all bit widths however.
4563 declare i32 @llvm.cttz.i8 (i8 <src>)
4564 declare i32 @llvm.cttz.i16(i16 <src>)
4565 declare i32 @llvm.cttz.i32(i32 <src>)
4566 declare i32 @llvm.cttz.i64(i64 <src>)
4567 declare i32 @llvm.cttz.i256(i256 <src>)
4573 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
4580 The only argument is the value to be counted. The argument may be of any
4581 integer type. The return type must match the argument type.
4587 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
4588 in a variable. If the src == 0 then the result is the size in bits of the type
4589 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
4593 <!-- _______________________________________________________________________ -->
4594 <div class="doc_subsubsection">
4595 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
4598 <div class="doc_text">
4601 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
4602 on any integer bit width.
4604 declare i17 @llvm.part.select.i17.i17 (i17 %val, i32 %loBit, i32 %hiBit)
4605 declare i29 @llvm.part.select.i29.i29 (i29 %val, i32 %loBit, i32 %hiBit)
4609 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
4610 range of bits from an integer value and returns them in the same bit width as
4611 the original value.</p>
4614 <p>The first argument, <tt>%val</tt> and the result may be integer types of
4615 any bit width but they must have the same bit width. The second and third
4616 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
4619 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
4620 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
4621 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
4622 operates in forward mode.</p>
4623 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
4624 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
4625 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
4627 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
4628 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
4629 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
4630 to determine the number of bits to retain.</li>
4631 <li>A mask of the retained bits is created by shifting a -1 value.</li>
4632 <li>The mask is ANDed with <tt>%val</tt> to produce the result.
4634 <p>In reverse mode, a similar computation is made except that:</p>
4636 <li>The bits selected wrap around to include both the highest and lowest bits.
4637 For example, part.select(i16 X, 4, 7) selects bits from X with a mask of
4638 0x00F0 (forwards case) while part.select(i16 X, 8, 3) selects bits from X
4639 with a mask of 0xFF0F.</li>
4640 <li>The bits returned in the reverse case are reversed. So, if X has the value
4641 0x6ACF and we apply part.select(i16 X, 8, 3) to it, we get back the value
4646 <div class="doc_subsubsection">
4647 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
4650 <div class="doc_text">
4653 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
4654 on any integer bit width.
4656 declare i17 @llvm.part.set.i17.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
4657 declare i29 @llvm.part.set.i29.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
4661 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
4662 of bits in an integer value with another integer value. It returns the integer
4663 with the replaced bits.</p>
4666 <p>The first argument, <tt>%val</tt> and the result may be integer types of
4667 any bit width but they must have the same bit width. <tt>%val</tt> is the value
4668 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
4669 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
4670 type since they specify only a bit index.</p>
4673 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
4674 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
4675 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
4676 operates in forward mode.</p>
4677 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
4678 truncating it down to the size of the replacement area or zero extending it
4679 up to that size.</p>
4680 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
4681 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
4682 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
4683 to the <tt>%hi</tt>th bit.
4684 <p>In reverse mode, a similar computation is made except that the bits replaced
4685 wrap around to include both the highest and lowest bits. For example, if a
4686 16 bit value is being replaced then <tt>%lo=8</tt> and <tt>%hi=4</tt> would
4687 cause these bits to be set: <tt>0xFF1F</tt>.</p>
4690 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
4691 llvm.part.set(0xFFFF, 0, 7, 4) -> 0x0060
4692 llvm.part.set(0xFFFF, 0, 8, 3) -> 0x00F0
4693 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
4697 <!-- ======================================================================= -->
4698 <div class="doc_subsection">
4699 <a name="int_debugger">Debugger Intrinsics</a>
4702 <div class="doc_text">
4704 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
4705 are described in the <a
4706 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
4707 Debugging</a> document.
4712 <!-- ======================================================================= -->
4713 <div class="doc_subsection">
4714 <a name="int_eh">Exception Handling Intrinsics</a>
4717 <div class="doc_text">
4718 <p> The LLVM exception handling intrinsics (which all start with
4719 <tt>llvm.eh.</tt> prefix), are described in the <a
4720 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
4721 Handling</a> document. </p>
4725 <!-- *********************************************************************** -->
4728 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
4729 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
4730 <a href="http://validator.w3.org/check/referer"><img
4731 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
4733 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
4734 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
4735 Last modified: $Date$